Parallel Standard Practices DS

Front cover
InfoSphere DataStage
Parallel Framework
Standard Practices
Develop highly efficient and scalable
information integration applications
Investigate, design, and develop

data flow jobs
Get guidelines for cost

effective performance
Julius Lerm
Paul Christensen
ibm.com/redbooks
International Technical Support Organization
InfoSphere DataStage: Parallel Framework

Standard Practices
September 2010
SG24-7830-00
Note: Before using this information and the product it supports, read the information in
“Notices” on page xiii.
First Edition (September 2010)
This edition applies to Version 8, Release 1 of IBM InfoSphere Information Server (5724-Q36)
and Version 9, Release 0, Modification 1 of IBM InfoSphere Master Data Management Server
(5724-V51), and Version 5.3.2 of RDP.
© Copyright International Business Machines Corporation 2010. All rights reserved.

Note to U.S. Government Users Restricted Rights -- Use, duplication or disclosure restricted by GSA ADP
Schedule Contract with IBM Corp.
Contents
Notices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
Trademarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
The team who wrote this book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii
Now you can become a published author, too! . . . . . . . . . . . . . . . . . . . . . . . . xix
Comments welcome. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xix
Stay connected to IBM Redbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xx
Chapter 1. Data integration with Information Server and DataStage . . . . . 1

1.1 Information Server 8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.1 Architecture and information tiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 IBM Information Management InfoSphere Services . . . . . . . . . . . . . . . . . . 5
1.3 Center of Excellence for Data Integration (CEDI) . . . . . . . . . . . . . . . . . . . . 6
1.4 Workshops for IBM InfoSphere DataStage . . . . . . . . . . . . . . . . . . . . . . . . 10
Chapter 2. Data integration overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.1 Job sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Job types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.1 Transformation jobs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.2 Hybrid jobs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.3 Provisioning jobs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Chapter 3. Standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.1 Directory structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.1.1 Metadata layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.1.2 Data, install, and project directory structures . . . . . . . . . . . . . . . . . . 23
3.1.3 Extending the DataStage project for external entities . . . . . . . . . . . . 24
3.1.4 File staging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 Naming conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2.1 Key attributes of the naming convention . . . . . . . . . . . . . . . . . . . . . . 32
3.2.2 Designer object layout. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2.3 Documentation and metadata capture . . . . . . . . . . . . . . . . . . . . . . . 34
3.2.4 Naming conventions by object type . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3 Documentation and annotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.4 Working with source code control systems . . . . . . . . . . . . . . . . . . . . . . . . 50
3.4.1 Source code control standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.4.2 Using object categorization standards . . . . . . . . . . . . . . . . . . . . . . . 51
3.4.3 Export to source code control system . . . . . . . . . . . . . . . . . . . . . . . . 51
© Copyright IBM Corp. 2010. All rights reserved. iii

Chapter 4. Job parameter and environment variable management . . . . . 55
4.1 DataStage environment variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.1.1 DataStage environment variable scope . . . . . . . . . . . . . . . . . . . . . . 57
4.1.2 Special values for DataStage environment variables . . . . . . . . . . . . 58
4.1.3 Environment variable settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.1.4 Migrating project-level environment variables . . . . . . . . . . . . . . . . . . 60
4.2 DataStage job parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.2.1 When to use parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.2.2 Parameter standard practices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.2.3 Specifying default parameter values . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.2.4 Parameter sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
Chapter 5. Development guidelines. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

5.1 Modular development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.2 Establishing job boundaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.3 Job design templates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.4 Default job design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.5 Parallel shared containers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.6 Error and reject record handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.6.1 Reject handling with the Sequential File stage . . . . . . . . . . . . . . . . . 75
5.6.2 Reject handling with the Lookup stage . . . . . . . . . . . . . . . . . . . . . . . 76
5.6.3 Reject handling with the Transformer stage . . . . . . . . . . . . . . . . . . . 77
5.6.4 Reject handling with Target Database stages. . . . . . . . . . . . . . . . . . 78
5.6.5 Error processing requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.7 Component usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.7.1 Server Edition components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.7.2 Copy stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.7.3 Parallel datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.7.4 Parallel Transformer stages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.7.5 BuildOp stages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.8 Job design considerations for usage and impact analysis . . . . . . . . . . . . 87
5.8.1 Maintaining JobDesign:Table definition connection . . . . . . . . . . . . . 88
5.8.2 Verifying the job design:table definition connection . . . . . . . . . . . . . 89
Chapter 6. Partitioning and collecting. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

6.1 Partition types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6.1.1 Auto partitioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
6.1.2 Keyless partitioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.1.3 Keyed partitioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.1.4 Hash partitioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.2 Monitoring partitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
6.3 Partition methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.4 Partitioning examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
iv InfoSphere DataStage: Parallel Framework Standard Practices

6.4.1 Partitioning example 1: Optimized partitioning . . . . . . . . . . . . . . . . 108
6.4.2 Partitioning example 2: Use of Entire partitioning . . . . . . . . . . . . . . 109
6.5 Collector types. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.5.1 Auto collector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.5.2 Round-robin collector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.5.3 Ordered collector. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
6.5.4 Sort Merge collector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
6.6 Collecting methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
Chapter 7. Sorting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

7.1 Partition and sort keys. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
7.2 Complete (Total) sort. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
7.3 Link sort and Sort stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
7.3.1 Link sort. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
7.3.2 Sort stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
7.4 Stable sort . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
7.5 Subsorts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
7.6 Automatically-inserted sorts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
7.7 Sort methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
7.8 Tuning sort . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
7.8.1 Sorts and variable-length fields. . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Chapter 8. File Stage usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

8.1 Dataset usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
8.2 Sequential File stages (Import and export) . . . . . . . . . . . . . . . . . . . . . . . 130
8.2.1 Reading from a sequential file in parallel . . . . . . . . . . . . . . . . . . . . 130
8.2.2 Writing to a sequential file in parallel. . . . . . . . . . . . . . . . . . . . . . . . 130
8.2.3 Separating I/O from column import . . . . . . . . . . . . . . . . . . . . . . . . . 131
8.2.4 Partitioning sequential file reads . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
8.2.5 Sequential file (Export) buffering . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
8.2.6 Parameterized sequential file format. . . . . . . . . . . . . . . . . . . . . . . . 132
8.2.7 Reading and writing nullable columns. . . . . . . . . . . . . . . . . . . . . . . 132
8.2.8 Reading from and writing to fixed-length files . . . . . . . . . . . . . . . . . 133
8.2.9 Reading bounded-length VARCHAR columns . . . . . . . . . . . . . . . . 134
8.2.10 Tuning sequential file performance . . . . . . . . . . . . . . . . . . . . . . . . 134
8.3 Complex Flat File stage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
8.3.1 CFF stage data type mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
8.4 Filesets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
Chapter 9. Transformation languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

9.1 Transformer stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
9.1.1 Transformer NULL handling and reject link. . . . . . . . . . . . . . . . . . . 140
9.1.2 Parallel Transformer system variables . . . . . . . . . . . . . . . . . . . . . . 141
9.1.3 Transformer derivation evaluation. . . . . . . . . . . . . . . . . . . . . . . . . . 141
Contents v
9.1.4 Conditionally aborting jobs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
9.1.5 Using environment variable parameters . . . . . . . . . . . . . . . . . . . . . 142
9.1.6 Transformer decimal arithmetic. . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
9.1.7 Optimizing Transformer expressions and stage variables . . . . . . . 143
9.2 Modify stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
9.2.1 Modify and null handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
9.2.2 Modify and string trim . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
9.3 Filter and Switch stages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
Chapter 10. Combining data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

10.1 Lookup versus Join versus Merge. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
10.2 Capturing unmatched records from a Join . . . . . . . . . . . . . . . . . . . . . . 150
10.3 The Aggregator stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
10.3.1 Aggregation method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
10.3.2 Aggregation data type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
10.3.3 Performing total aggregations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
10.4 Comparison stages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
10.5 Checksum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
10.6 SCD stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
Chapter 11. Restructuring data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

11.1 Complex data types. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
11.1.1 Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
11.1.2 Subrecords . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
11.1.3 Tagged fields. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
11.2 The Restructure library . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
11.2.1 Tagbatch and Tagswitch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
11.2.2 Importing complex record types . . . . . . . . . . . . . . . . . . . . . . . . . . 169
11.3 The Pivot Enterprise stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
Chapter 12. Performance tuning job designs . . . . . . . . . . . . . . . . . . . . . . 173

12.1 Designing a job for optimal performance. . . . . . . . . . . . . . . . . . . . . . . . 174
12.2 Understanding operator combination . . . . . . . . . . . . . . . . . . . . . . . . . . 176
12.3 Minimizing runtime processes and resource
requirements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
12.4 Understanding buffering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
12.4.1 Inter-operator transport buffering . . . . . . . . . . . . . . . . . . . . . . . . . 180
12.4.2 Deadlock prevention buffering . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
Chapter 13. Database stage guidelines. . . . . . . . . . . . . . . . . . . . . . . . . . . 189

13.1 Existing database development overview . . . . . . . . . . . . . . . . . . . . . . . 190
13.1.1 Existing database stage types . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
13.1.2 Database metadata. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
13.1.3 Optimizing select lists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
vi InfoSphere DataStage: Parallel Framework Standard Practices

13.1.4 Testing database connectivity. . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
13.1.5 Designing for restart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
13.1.6 Database OPEN and CLOSE commands . . . . . . . . . . . . . . . . . . . 196
13.1.7 Database sparse lookup versus join . . . . . . . . . . . . . . . . . . . . . . . 197
13.1.8 Appropriate use of SQL and DataStage . . . . . . . . . . . . . . . . . . . . 197
13.2 Existing DB2 guidelines. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
13.2.1 Existing DB2 stage types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
13.2.2 Connecting to DB2 with the DB2/UDB Enterprise stage . . . . . . . . 200
13.2.3 Configuring DB2 multiple instances in one DataStage job . . . . . . 201
13.2.4 DB2/UDB Enterprise stage column names . . . . . . . . . . . . . . . . . . 202
13.2.5 DB2/API stage column names . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
13.2.6 DB2/UDB Enterprise stage data type mapping . . . . . . . . . . . . . . . 203
13.2.7 DB2/UDB Enterprise stage options . . . . . . . . . . . . . . . . . . . . . . . . 205
13.2.8 Performance notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
13.3 Existing Informix database guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . 206
13.3.1 Informix Enterprise stage column names . . . . . . . . . . . . . . . . . . . 206
13.3.2 Informix Enterprise stage data type mapping . . . . . . . . . . . . . . . . 206
13.4 ODBC Enterprise guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
13.4.1 ODBC Enterprise stage column names . . . . . . . . . . . . . . . . . . . . 208
13.4.2 ODBC Enterprise stage data type mapping . . . . . . . . . . . . . . . . . 208
13.4.3 Reading ODBC sources in parallel . . . . . . . . . . . . . . . . . . . . . . . . 209
13.4.4 Writing to ODBC targets in parallel . . . . . . . . . . . . . . . . . . . . . . . . 209
13.5 Oracle database guidelines. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
13.5.1 Oracle Enterprise stage column names . . . . . . . . . . . . . . . . . . . . 210
13.5.2 Oracle Enterprise stage data type mapping . . . . . . . . . . . . . . . . . 210
13.5.3 Reading from Oracle in parallel. . . . . . . . . . . . . . . . . . . . . . . . . . . 211
13.5.4 Oracle load options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
13.6 Sybase Enterprise guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
13.6.1 Sybase Enterprise stage column names. . . . . . . . . . . . . . . . . . . . 212
13.6.2 Sybase Enterprise stage data type mapping . . . . . . . . . . . . . . . . 212
13.7 Existing Teradata database guidelines . . . . . . . . . . . . . . . . . . . . . . . . . 214
13.7.1 Choosing the proper Teradata stage . . . . . . . . . . . . . . . . . . . . . . 214
13.7.2 Source Teradata stages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
13.7.3 Target Teradata stages. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
13.7.4 Teradata Enterprise stage column names . . . . . . . . . . . . . . . . . . 216
13.7.5 Teradata Enterprise stage data type mapping . . . . . . . . . . . . . . . 216
13.7.6 Specifying Teradata passwords with special characters. . . . . . . . 217
13.7.7 Teradata Enterprise settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
13.7.8 Improving Teradata Enterprise performance . . . . . . . . . . . . . . . . 218
13.8 Netezza Enterprise stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
13.8.1 Netezza write methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
13.8.2 Limitations of Netezza Write stage . . . . . . . . . . . . . . . . . . . . . . . . 219
13.8.3 Netezza Enterprise error logs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
Contents vii
Chapter 14. Connector stage guidelines. . . . . . . . . . . . . . . . . . . . . . . . . . 221
14.1 Connectors and the connector framework . . . . . . . . . . . . . . . . . . . . . . 222
14.1.1 Connectors in parallel jobs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
14.1.2 Large object (LOB) support. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
14.1.3 Reject Links. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
14.1.4 Schema reconciliation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
14.1.5 Stage editor concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
14.1.6 Connection objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
14.1.7 SQL Builder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
14.1.8 Metadata importation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
14.2 ODBC Connector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
14.3 WebSphere MQ Connector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
14.4 Teradata Connector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
14.4.1 Teradata Connector advantages. . . . . . . . . . . . . . . . . . . . . . . . . . 237
14.4.2 Parallel Synchronization Table . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
14.4.3 Parallel Transport operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
14.4.4 Cleanup after an aborted load or update . . . . . . . . . . . . . . . . . . . 238
14.4.5 Environment variables for debugging job execution . . . . . . . . . . . 239
14.4.6 Comparison with existing Teradata stages . . . . . . . . . . . . . . . . . . 239
14.5 DB2 Connector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
14.5.1 New features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
14.5.2 Using rejects with user-defined SQL . . . . . . . . . . . . . . . . . . . . . . . 244
14.5.3 Using alternate conductor setting . . . . . . . . . . . . . . . . . . . . . . . . . 245
14.5.4 Comparison with existing DB2 stages. . . . . . . . . . . . . . . . . . . . . . 246
14.6 Oracle Connector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
14.6.1 New features and improvements . . . . . . . . . . . . . . . . . . . . . . . . . 251
14.6.2 Comparison with Oracle Enterprise . . . . . . . . . . . . . . . . . . . . . . . 252
14.7 DT stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
14.8 SalesForce Connector. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
14.9 Essbase connector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
14.10 SWG Connector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
Chapter 15. Batch data flow design. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259

15.1 High performance batch data flow design goals . . . . . . . . . . . . . . . . . . 260
15.1.1 Minimize time required to complete batch processing . . . . . . . . . 260
15.1.2 Build scalable jobs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
15.1.3 Minimize the impact of startup time . . . . . . . . . . . . . . . . . . . . . . . 260
15.1.4 Optimize network, I/O and memory usage . . . . . . . . . . . . . . . . . . 261
15.1.5 Plan job concurrency and degrees of parallelism . . . . . . . . . . . . . 262
15.2 Common bad patterns. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
15.2.1 DS server mentality for parallel jobs . . . . . . . . . . . . . . . . . . . . . . . 262
15.2.2 Database sparse lookups. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
15.2.3 Processing full source database refreshes . . . . . . . . . . . . . . . . . . 264
viii InfoSphere DataStage: Parallel Framework Standard Practices

15.2.4 Extracting much and using little (reference datasets) . . . . . . . . . . 264
15.2.5 Reference data is too large to fit into physical memory . . . . . . . . 265
15.2.6 Loading and re-extracting the same data . . . . . . . . . . . . . . . . . . . 265
15.2.7 One sequence run per input/output file . . . . . . . . . . . . . . . . . . . . . 265
15.3 Optimal number of stages per job . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
15.4 Checkpoint/Restart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
15.5 Balanced optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
15.5.1 Transformations inside the database . . . . . . . . . . . . . . . . . . . . . . 268
15.5.2 Transformations with DataStage . . . . . . . . . . . . . . . . . . . . . . . . . . 268
15.6 Batch data flow patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
15.6.1 Restricting incoming data from the source . . . . . . . . . . . . . . . . . . 270
15.6.2 A fundamental problem: Reference lookup resolution . . . . . . . . . 271
15.6.3 A sample database model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
15.6.4 Restricting the reference lookup dataset. . . . . . . . . . . . . . . . . . . . 273
15.6.5 Correlating data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
15.6.6 Keeping information server as the transformation hub . . . . . . . . 281
15.6.7 Accumulating reference data in local datasets . . . . . . . . . . . . . . . 282
15.6.8 Minimize number of sequence runs per processing window. . . . . 285
15.6.9 Separating database interfacing and transformation jobs . . . . . . . 286
15.6.10 Extracting data efficiently . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
15.6.11 Uploading data efficiently . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
Chapter 16. Real-time data flow design. . . . . . . . . . . . . . . . . . . . . . . . . . . 293

16.1 Definition of real-time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
16.2 Mini-batch approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
16.3 Parallel framework in real-time applications . . . . . . . . . . . . . . . . . . . . . 297
16.4 DataStage extensions for real-time applications . . . . . . . . . . . . . . . . . . 299
16.4.1 Always-on source stage types . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
16.4.2 End-of-wave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
16.4.3 Transaction support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
16.5 Job topologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
16.5.1 Summary of stage usage guidelines . . . . . . . . . . . . . . . . . . . . . . . 312
16.5.2 ISD batch topologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
16.6 MQConnector/DTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
16.6.1 Aspects of DTS application development . . . . . . . . . . . . . . . . . . . 314
16.6.2 Reference documentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
16.6.3 A sample basic DTS job . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
16.6.4 Design topology rules for DTS jobs. . . . . . . . . . . . . . . . . . . . . . . . 316
16.6.5 Transactional processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
16.6.6 MQ/DTS and the Information Server Framework . . . . . . . . . . . . . 319
16.6.7 Sample job and basic properties. . . . . . . . . . . . . . . . . . . . . . . . . . 321
16.6.8 Runtime Topologies for DTS jobs . . . . . . . . . . . . . . . . . . . . . . . . . 326
16.6.9 Processing order of input links . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
Contents ix
16.6.10 Rejecting messages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
16.6.11 Database contention . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
16.6.12 Scalability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
16.6.13 Design patterns to avoid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
16.7 InfoSphere Information Services Director . . . . . . . . . . . . . . . . . . . . . . . 346
16.7.1 The scope of this section . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
16.7.2 Design topology rules for always-on ISD jobs. . . . . . . . . . . . . . . . 351
16.7.3 Scalability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
16.7.4 Synchronizing database stages with ISD output . . . . . . . . . . . . . 353
16.7.5 ISD with DTS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354
16.7.6 ISD with connectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
16.7.7 Re-partitioning in ISD jobs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
16.7.8 General considerations for using ISD jobs . . . . . . . . . . . . . . . . . . 359
16.7.9 Selecting server or EE jobs for publication through ISD . . . . . . . . 361
16.8 Transactional support in message-oriented applications . . . . . . . . . . . 362
16.9 Payload processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
16.10 Pipeline Parallelism challenges. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
16.10.1 Key collisions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 366
16.10.2 Data stubbing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
16.10.3 Parent/Child processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
16.11 Special custom plug-ins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
16.12 Special considerations for QualityStage . . . . . . . . . . . . . . . . . . . . . . . 373
Appendix A. Runtime topologies for distributed transaction jobs. . . . . 375

A.1 No ordering, no relationships . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376
A.2 No ordering, with relationships . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
A.3 Bypassing work queues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
Appendix B. Standard practices summary . . . . . . . . . . . . . . . . . . . . . . . . 381

B.1 Standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
Directory Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
Data, install, and project directory structure . . . . . . . . . . . . . . . . . . . . . . . 383
B.2 Development guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
B.3 Component usage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
B.4 DataStage data types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
B.5 Partitioning data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
B.6 Collecting data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386
B.7 Sorting. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
B.8 Stage-specific guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
B.9 Database stage guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 388
B.10 Troubleshooting and monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
Appendix C. DataStage naming reference . . . . . . . . . . . . . . . . . . . . . . . . 391
x InfoSphere DataStage: Parallel Framework Standard Practices

Appendix D. Example job template . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
Appendix E. Understanding the parallel job score . . . . . . . . . . . . . . . . . 401

E.1 Viewing the job score . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402
E.2 Parallel job score components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
E.2.1 Job Score: Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
E.2.2 Job Score: Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
Appendix F. Estimating the size of a parallel dataset . . . . . . . . . . . . . . . 411
Appendix G. Environment variables reference. . . . . . . . . . . . . . . . . . . . . 413
Appendix H. DataStage data types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423

H.1 Parallel data types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
H.2 Null handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
H.3 Runtime column propagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429
Related publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431

IBM Redbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
Other publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
Online resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
How to get Redbooks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432
Help from IBM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432
Contents xi
xii InfoSphere DataStage: Parallel Framework Standard Practices
Notices
This information was developed for products and services offered in the U.S.A.
IBM may not offer the products, services, or features discussed in this document in other countries. Consult
your local IBM representative for information on the products and services currently available in your area.
Any reference to an IBM product, program, or service is not intended to state or imply that only that IBM
product, program, or service may be used. Any functionally equivalent product, program, or service that
does not infringe any IBM intellectual property right may be used instead. However, it is the user's
responsibility to evaluate and verify the operation of any non-IBM product, program, or service.
IBM may have patents or pending patent applications covering subject matter described in this document.
The furnishing of this document does not give you any license to these patents. You can send license
inquiries, in writing, to:
IBM Director of Licensing, IBM Corporation, North Castle Drive, Armonk, NY 10504-1785 U.S.A.
The following paragraph does not apply to the United Kingdom or any other country where such
provisions are inconsistent with local law: INTERNATIONAL BUSINESS MACHINES CORPORATION
PROVIDES THIS PUBLICATION "AS IS" WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESS OR
IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF NON-INFRINGEMENT,
MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Some states do not allow disclaimer
of express or implied warranties in certain transactions, therefore, this statement may not apply to you.
This information could include technical inaccuracies or typographical errors. Changes are periodically made
to the information herein; these changes will be incorporated in new editions of the publication. IBM may
make improvements and/or changes in the product(s) and/or the program(s) described in this publication at
any time without notice.
Any references in this information to non-IBM Web sites are provided for convenience only and do not in any
manner serve as an endorsement of those Web sites. The materials at those Web sites are not part of the
materials for this IBM product and use of those Web sites is at your own risk.
IBM may use or distribute any of the information you supply in any way it believes appropriate without
incurring any obligation to you.
Information concerning non-IBM products was obtained from the suppliers of those products, their published
announcements or other publicly available sources. IBM has not tested those products and cannot confirm
the accuracy of performance, compatibility or any other claims related to non-IBM products. Questions on
the capabilities of non-IBM products should be addressed to the suppliers of those products.
This information contains examples of data and reports used in daily business operations. To illustrate them
as completely as possible, the examples include the names of individuals, companies, brands, and products.
All of these names are fictitious and any similarity to the names and addresses used by an actual business
enterprise is entirely coincidental.
COPYRIGHT LICENSE:
This information contains sample application programs in source language, which illustrate programming
techniques on various operating platforms. You may copy, modify, and distribute these sample programs in
any form without payment to IBM, for the purposes of developing, using, marketing or distributing application
programs conforming to the application programming interface for the operating platform for which the
sample programs are written. These examples have not been thoroughly tested under all conditions. IBM,
therefore, cannot guarantee or imply reliability, serviceability, or function of these programs.
© Copyright IBM Corp. 2010. All rights reserved. xiii

Trademarks
IBM, the IBM logo, and ibm.com are trademarks or registered trademarks of International Business
Machines Corporation in the United States, other countries, or both. These and other IBM trademarked
terms are marked on their first occurrence in this information with the appropriate symbol (® or ™),
indicating US registered or common law trademarks owned by IBM at the time this information was
published. Such trademarks may also be registered or common law trademarks in other countries. A current
list of IBM trademarks is available on the Web at http://www.ibm.com/legal/copytrade.shtml
The following terms are trademarks of the International Business Machines Corporation in the United States,
other countries, or both:
AIX® Information Agenda™ MVS™

ClearCase® Informix® Orchestrate®
DataStage® InfoSphere™ Redbooks®
DB2® Iterations® Redbooks (logo) ®
IBM® MQSeries® WebSphere®
The following terms are trademarks of other companies:
Java, and all Java-based trademarks are trademarks of Sun Microsystems, Inc. in the United States, other
countries, or both.
Windows, and the Windows logo are trademarks of Microsoft Corporation in the United States, other
countries, or both.
UNIX is a registered trademark of The Open Group in the United States and other countries.
Linux is a trademark of Linus Torvalds in the United States, other countries, or both.
Other company, product, or service names may be trademarks or service marks of others.
xiv InfoSphere DataStage: Parallel Framework Standard Practices

Preface
In this IBM® Redbooks® publication, we present guidelines for the development

of highly efficient and scalable information integration applications with
InfoSphere™ DataStage® (DS) parallel jobs.
InfoSphere DataStage is at the core of IBM Information Server, providing

components that yield a high degree of freedom. For any particular problem
there might be multiple solutions, which tend to be influenced by personal
preferences, background, and previous experience. All too often, those solutions
yield less than optimal, and non-scalable, implementations.
This book includes a comprehensive detailed description of the components

available, and descriptions on how to use them to obtain scalable and efficient
solutions, for both batch and real-time scenarios.
The advice provided in this document is the result of the combined proven
experience from a number of expert practitioners in the field of high performance
information integration, evolved over several years.
This book is intended for IT architects, Information Management specialists, and

Information Integration specialists responsible for delivering cost-effective IBM
InfoSphere DataStage performance on all platforms.
This book is organized as follows:

򐂰 Chapter 1, “Data integration with Information Server and DataStage” on
page 1
This chapter presents an overview of the Information Integration components
and services offered by IBM Information Management. This includes a
discussion of Information Server 8 and its layers (client, services, metadata,
and engine).
򐂰 Chapter 2, “Data integration overview” on page 11
In this chapter we describe the use of job sequences and other types of jobs
commonly found in data integration applications.
򐂰 Chapter 3, “Standards” on page 21
In chapter 3, we describe and discuss recommendations for the adoption of
the following development and deployment standards:
– Directory structures
– Naming conventions
© Copyright IBM Corp. 2010. All rights reserved. xv

– Graphical job layouts
– source control systems
– categorization standards
򐂰 Chapter 4, “Job parameter and environment variable management” on
page 55
In this chapter we describe how to use job parameters, parameter sets, and
environment variables.
򐂰 Chapter 5, “Development guidelines” on page 69
In this chapter, we discuss development guidelines for modular designs,
adequate error handling, and how to facilitate metadata management.
򐂰 Chapters 6–14, beginning on page 91
In this sequence of chapters of the book, we provide detailed description of a
number of the features and related stage types of DataStage. The following
list details the features covered:
– Partitioning and collecting
– Sorting
– File stage Usage
– Transformation Languages
– Combining Data
– Restructuring Data
– Performance Tuning Job Designs
– Existing Database stage Guidelines
– Connector stage Guidelines
򐂰 Chapter 15, “Batch data flow design” on page 259
In this chapter we present recommendations on how to obtain efficient and
scalable job designs to process large amounts of data in batch mode. We also
list the primary design goals for batch applications and discuss a number of
common bad design patterns.
The emphasis is on proven job design patterns that apply to most, if not all,
large volume batch processing applications.
򐂰 Chapter 16, “Real-time data flow design” on page 293
In this chapter we present recommendations on how to obtain efficient and
scalable job designs to process large amounts of data in real-time mode.
Though partitioning and pipeline parallelism enables scalability and efficiency,
it introduces challenges in real-time scenarios. We present a comprehensive
discussion on what real-time means in the context of InfoSphere DataStage:
– Job topologies
– Message-oriented processing with MQSeries®
– SOA applications with Information Services Director (ISD)
xvi InfoSphere DataStage: Parallel Framework Standard Practices

– Real-time scalability
In addition, we present techniques to overcome those challenges introduced
by parallelism.
򐂰 Appendices, beginning on page 375
We provide a number of supporting topics in the appendix:
– Runtime topologies for distributed transaction jobs
– Standard practices summary
– DataStage naming reference
– Example job template
– Understanding the parallel job score
– Estimating the size of a parallel dataset
– Environmental variables reference
– DataStage data types
Document history
Prior to this Redbooks publication, there was DataStage release documentation
made available dated April 15, 2009, written by Julius Lerm. This book updates
and extends the information in that initial release documentation with
terminology, import/export mechanisms, and job parameter handling (including
parameter sets).
The team who wrote this book

This book was produced by the following authors, along with contributions from a
number of their colleagues. The authors are listed, along with a short
biographical sketch of each.
Julius Lerm is a Technical Architect, and a member of the Center of Excellence

of IBM Information Management, Analytics & Optimization Software Services. He
has 16 years experience in solution architecting, and developing and
performance tuning of large scale parallel database and information integration
applications. His experience includes extensive development of custom
components and extensions to the DataStage parallel framework, as well as
tools that bridge product functionality gaps and address complex integration
problems. He has provided guidance, mentoring, and training to customers
worldwide in transactional, data warehousing, and service-oriented and
message-based processing projects. He has also presented at several
conferences. Julius holds Bachelor’s and Master’s degrees in Computer Science
from Federal University of Rio Grande do Sul (UFRGS, Brazil).
Preface xvii
Paul Christensen is a Technical Architect and member of the worldwide IBM
Information Agenda™ Architecture team. With 19 years of experience in
enterprise data management and parallel computing technologies, he has led
the successful design, implementation, and management of large-scale Data
Integration and Information Management solutions using the IBM Information
Agenda and partner portfolios. Paul's experience includes early hardware-based
parallel computing platforms, massively parallel databases including Informix®
and DB2®, and the parallel framework of IBM Information Server and DataStage.
To facilitate successful customer and partner deployments using IBM Information
Server, he has helped to develop standard practices, course material, and
technical certifications. Paul holds a Bachelor’s degree in Electrical Engineering
from Drexel University, and is an IBM Certified Solution Developer.
Other Contributors
We would like to give special thanks to the following contributing authors whose
input added significant value to this publication.
Mike Carney - Technical Architect, IBM Software Group, Information
Management, Westford, MA
Tony Curcio - DataStage Product Manager, IBM Software Group, Information
Management, Charlotte, NC
Patrick Owen - Technical Architect, IBM Software Group, Information
Management, Little Rock, AR
Steve Rigo - Technical Architect, IBM Software Group, Information Management,
Atlanta, GA
Ernie Ostic - Technical Sales Specialist, IBM Software Group, Worldwide Sales,
Newark, NJ
Paul Stanley - Product Development Engineer, IBM Software Group, Information
Management, Boca Raton, FL
In the following sections we thank others who have contributed to the
development and publication of this IBM Redbooks publication.
From IBM Locations Worldwide
Tim Davis - Executive Director, Information Agenda Architecture Group, IBM
Software Group, Information Management, Littleton, MA
Susan Laime - IM Analytics and Optimization Software Services, IBM Software
Group, Information Management, Littleton, MA
Margaret Noel - Integration Architect, IBM Software Group, Information
Management, Atlantic Beach, FL
From the International Technical Support Organization

Chuck Ballard - Project Manager at the International Technical Support
organization, in San Jose, California. Chuck managed the processes
required to format and publish this IBM Redbooks Publication.
Mary Comianos - Publications Management
Emma Jacobs - Graphics
James Hoy - Editor
xviii InfoSphere DataStage: Parallel Framework Standard Practices

Now you can become a published author, too!
Here's an opportunity to spotlight your skills, grow your career, and become a
published author - all at the same time! Join an ITSO residency project and help
write a book in your area of expertise, while honing your experience using
leading-edge technologies. Your efforts will help to increase product acceptance
and customer satisfaction, as you expand your network of technical contacts and
relationships. Residencies run from two to six weeks in length, and you can
participate either in person or as a remote resident working from your home
base.
Find out more about the residency program, browse the residency index, and
apply online at:
ibm.com/redbooks/residencies.html
Comments welcome
Your comments are important to us!
We want our books to be as helpful as possible. Send us your comments about

this book or other IBM Redbooks publications in one of the following ways:
򐂰 Use the online Contact us review Redbooks form found at:
ibm.com/redbooks
򐂰 Send your comments in an e-mail to:
redbooks@us.ibm.com
򐂰 Mail your comments to:
IBM Corporation, International Technical Support Organization
Dept. HYTD Mail Station P099
2455 South Road
Poughkeepsie, NY 12601-5400
Preface xix
Stay connected to IBM Redbooks
򐂰 Find us on Facebook:
http://www.facebook.com/pages/IBM-Redbooks/178023492563?ref=ts
򐂰 Follow us on twitter:
http://twitter.com/ibmredbooks
򐂰 Look for us on LinkedIn:
http://www.linkedin.com/groups?home=&gid=2130806
򐂰 Explore new Redbooks publications, residencies, and workshops with the
IBM Redbooks weekly newsletter:
https://www.redbooks.ibm.com/Redbooks.nsf/subscribe?OpenForm
򐂰 Stay current on recent Redbooks publications with RSS Feeds:
http://www.redbooks.ibm.com/rss.html
xx InfoSphere DataStage: Parallel Framework Standard Practices

1
Chapter 1. Data integration with

Information Server and
DataStage
In this chapter we discuss and describe the concepts, education, and services
that are available to help you get started with your data integration activities.
Information integration is the process of integrating and transforming data and
content to deliver authoritative, consistent, timely and complete information, and
governing its quality throughout its life cycle. Core to those activities is the
InfoSphere Information Server platform, and InfoSphere DataStage (DS).
IBM InfoSphere Information Server is a software platform that helps

organizations derive more value from the complex, heterogeneous information
spread across their systems. It provides breakthrough collaboration, productivity,
and performance for cleansing, transforming, and moving this information
consistently and securely throughout the enterprise. It can then be accessed and
used in new ways to drive innovation, increase operational efficiency, and lower
risk
IBM InfoSphere DataStage integrates data across multiple and high volumes
data sources and target applications. It integrates data on demand with a high
performance parallel framework, extended metadata management, and
© Copyright IBM Corp. 2010. All rights reserved. 1

enterprise connectivity. DataStage supports the collection, integration, and
transformation of large volumes of data, with data structures ranging from simple
to highly complex.
DataStage can manage data arriving in real-time as well as data received on a

periodic or scheduled basis. This enables companies to solve large-scale
business problems through high-performance processing of massive data
volumes. By making use of the parallel processing capabilities of multiprocessor
hardware platforms, IBM InfoSphere DataStage Enterprise Edition can scale to
satisfy the demands of ever-growing data volumes, stringent real-time
requirements, and ever-shrinking batch windows.
Along with these key components, establishing consistent development

standards helps to improve developer productivity and reduce ongoing
maintenance costs. Development standards can also make it easier to integrate
external processes (such as automated auditing and reporting) and to build
technical and support documentation.
With these components and a great set of standard practices, you are on your
way to a highly successful data integration effort. To help you further along the
way, this book also provides a brief overview of a number of services and
education offerings by IBM.
But first, to aid in understanding, we provide a brief overview of the InfoSphere

Information Server 8.
2 InfoSphere DataStage: Parallel Framework Standard Practices

1.1 Information Server 8
Information Server 8 implements a new architecture that differs from earlier
versions of DataStage.
DataStage 7.X consisted of a two-tier infrastructure, with clients connected

directly to the DSEngine. The DSEngine stored all of the metadata and runtime
information, as well as controlled the execution of jobs.
IBM InfoSphere Information Server 8 is installed in layers that are mapped to the
physical hardware. In addition to the main product modules, product components
are installed in each tier as needed.
1.1.1 Architecture and information tiers

This new architecture does not affect the way DataStage jobs are developed.
The DataStage parallel framework remains the same, with a few minimal
changes to internal mechanisms that do not impact job designs in any way.
From a job design perspective, the product has interesting new features:
򐂰 New stages, such as Database Connectors, Slowly Changing Dimensions,
and Distributed Transaction.
򐂰 Job Parameter Sets
򐂰 Balanced optimization, a capability that automatically or semi-automatically
rewrites jobs to make use of RDBMS capabilities for transformations.
Information Server also provides new features for job developers and
administrators, such as a more powerful import/export facility, a job comparison
tool, and an impact analysis tool.
Chapter 1. Data integration with Information Server and DataStage 3

The information tiers work together to provide services, job execution, metadata,
and other storage, as shown in Figure 1-1.
Client Tier
Services Tier Engine Tier
Services for IBM IBM Information

Information Server Server Engine
Common Services Connectors Packs
InfoSphere QualityStage
Product-specific services
modules
Service Agents
Application Server (Logging, Communications (ASB),
Job Monitor, Resource Tracker)
Metadata
Repository Tier
Figure 1-1 Information Server tiers

The following list describes the information tiers shown in Figure 1-1:
򐂰 Client
Product module clients that are not Web-based and that are used for
development and administration in IBM InfoSphere Information Server. The
client tier includes the IBM InfoSphere Information Server console, IBM
InfoSphere DataStage and QualityStage clients, and other clients.
򐂰 Engine
Runtime engine that runs jobs and other tasks for product modules that
require the engine.
򐂰 Metadata repository
Database that stores the shared metadata, data, and configuration for IBM
InfoSphere Information Server and the product modules.
򐂰 Services
Common and product-specific services for IBM InfoSphere Information
Server along with IBM WebSphere® Application Server (application server).
In this document we focus on aspects related to parallel job development, which

is directly related to the Client and Engine layers.

1.2 IBM Information Management InfoSphere Services
IBM Information Management InfoSphere Professional Services offers a broad
range of workshops and services designed to help you achieve success in the
design, implementation, and rollout of critical information integration projects. An
overview of this range of services is depicted in Figure 1-2.
Figure 1-2 IBM InfoSphere Services overview

Based on that overview, we have listed a number of services offerings and their
description in Table 1-1.
Table 1-1 Services description

Services Description
offerings
Staff Whether through workshop delivery, project leadership, or mentored augmentation, the
Augmentation Professional Services staff of IBM Information Platform and Solutions use IBM
and Mentoring methodologies, standard practices, and experience developed through thousands of
successful engagements in a wide range of industries and government entities.
Learning IBM offers a variety of courses covering the IBM Information Management product
Services portfolio. The IBM blended learning approach is based on the principle that people learn
best when provided with a variety of learning methods that build upon and complement
each other. With that in mind, courses are delivered through a variety of mechanisms:
classroom, on-site, and Web-enabled FlexLearning.
Certification IBM offers a number of professional certifications through independent testing centers
worldwide. These certification exams provide a reliable, valid, and fair method of
assessing product skills and knowledge gained through classroom and real-world
experience.
Client Support IBM is committed to providing our customers with reliable technical support worldwide.
Services All client support services are available to customers who are covered under an active
IBM InfoSphere maintenance agreement. Our worldwide support organization is
dedicated to assuring your continued success with IBM InfoSphere products and
solutions.
Virtual The low cost Virtual Services offering is designed to supplement the global IBM
Services InfoSphere delivery team, as needed, by providing real-time, remote consulting
services. Virtual Services has a large pool of experienced resources that can provide IT
consulting, development, migration, and training services to customers for IBM
InfoSphere DataStage.
1.3 Center of Excellence for Data Integration (CEDI)

Establishing a CEDI in your enterprise can increase efficiency and drive down
the cost of implementing data integration projects. A CEDI can be responsible for
competency, readiness, accelerated mentored learning, common business rules,
standard practices, repeatable processes, and the development of custom
methods and components tailored to your business.

IBM InfoSphere Professional Services offerings can be delivered as part of a
strategic CEDI initiative, or on an as-needed basis across a project life cycle. The
IBM InfoSphere services offerings in an information integration project life cycle
are illustrated in Figure 1-3.
Installation and Configuration

Information Analysis
Information Exchange Data Flow and Job Design Standard Practices
and Discovery Data Quality Management Standard Practices
Strategic Test & Monitor &

Identify Startup Analysis Build
Planning Implement Refine
Requirements Definition, Administration, Management, and

Architecture, and Project Production Automation & Design
Planning
Iterations 2 Sizing and Capacity Planning
Grid Computing Discovery, Architecture, and Planning

Grid Computing Installation and Deployment
Figure 1-3 InfoSphere Services offerings and project life cycle
Table 1-2 lists the workshop offerings and descriptions for project startup.
Table 1-2 Project startup workshops

Project Startup Description
Workshops
Information Exchange Targeted for clients new to the IBM InfoSphere product portfolio, this workshop
and Discovery provides high-level recommendations on how to solve a customer’s particular
Workshop problem. IBM analyzes the data integration challenges outlined by the client and
develops a strategic approach for addressing those challenges.
Requirements Guiding clients through the critical process of establishing a framework for a
Definition, successful future project implementation, this workshop delivers a detailed
Architecture, and project plan, as well as a project blueprint. These deliverables document project
Project Planning parameters, current and conceptual end states, network topology, and data
Workshop architecture and hardware and software specifications. It also outlines a
communication plan, defines scope, and captures identified project risk.
Iterations® 2 IBM Iterations 2 is a framework for managing enterprise data integration projects
that integrate with existing customer methodologies. Iterations 2 is a
comprehensive, iterative, step-by-step approach that leads project teams from
initial planning and strategy through tactical implementation. This workshop
includes the Iterations 2 software, along with customized mentoring.

The workshop offerings and descriptions for standard practices are shown in
Table 1-3.
Table 1-3 Standard practices workshops

Standard Practices Description
Workshops
Installation and This workshop establishes a documented, repeatable process for installation and
Configuration configuration of IBM InfoSphere Information Server components. This might
Workshop involve review and validation of one or more existing Information Server
environments, or planning, performing, and documenting a new installation.
Information Analysis This workshop provides clients with a set of standard practices and a repeatable
Workshop methodology for analyzing the content, structure, and quality of data sources
using a combination of IBM InfoSphere Information Analyzer, QualityStage, and
Audit stage.
Data Flow and Job This workshop helps clients establish standards and templates for the design and
Design Standard development of parallel jobs using IBM InfoSphere DataStage through
Practices Workshop practitioner-led application of IBM standard practices to a client’s environment,
business, and technical requirements. The delivery includes a customized
standards document as well as custom job designs and templates for a focused
subject area.
Data Quality This workshop provides clients with a set of standard processes for the design
Management and development of data standardization, matching, and survivorship processes
Standard Practices using IBM InfoSphere QualityStage The data quality strategy formulates an
Workshop auditing and monitoring program to ensure on-going confidence in data accuracy,
consistency, and identification through client mentoring and sharing of IBM
standard practices.
Administration, This workshop provides customers with a customized tool kit and a set of proven
Management, and standard practices for integrating IBM InfoSphere DataStage into a client’s
Production existing production infrastructure (monitoring, scheduling, auditing/logging,
Automation change management) and for administering, managing and operating DataStage
Workshop environments.

The workshop offerings and descriptions for advanced deployment are shown in
Table 1-4.
Table 1-4 Advanced deployment workshops

Advanced Description
Deployment
Workshops
Health Check This workshop is targeted for clients currently engaged in IBM InfoSphere
Evaluation development efforts that are not progressing according to plan, or for clients seeking
validation of proposed plans prior to the commencement of new projects. It provides
review of, and recommendations for, core Extract Transform and Load (ETL)
development and operational environments by an IBM expert practitioner.
Sizing and The workshop provides clients with an action plan and set of recommendations for
Capacity Planning meeting current and future capacity requirements for data integration. This strategy
Workshop is based on analysis of business and technical requirements, data volumes and
growth projections, existing standards and technical architecture, and existing and
future data integration projects.
Performance This workshop guides a client’s technical staff through IBM standard practices and
Tuning Workshop methodologies for review, analysis, and performance optimization using a targeted
sample of client jobs and environments. This workshop identifies potential areas of
improvement, demonstrates IBM processes and techniques, and provides a final
report with recommended performance modifications and IBM performance tuning
guidelines.
High-Availability Using IBM InfoSphere Standard Practices for high availability, this workshop
Architecture presents a plan for meeting a customer’s high availability requirements using the
Workshop parallel framework of IBM InfoSphere DataStage. It implements the architectural
modifications necessary for high availability computing.
Grid Computing This workshop teaches the planning and readiness efforts required to support a
Discovery, future deployment of the parallel framework of IBM InfoSphere DataStage on Grid
Architecture and computing platforms. This workshop prepares the foundation on which a follow-on
Planning grid installation and deployment is executed, and includes hardware and software
Workshop recommendations and estimated scope.
Grid Computing In this workshop, the attendee installs, configures, and deploys the IBM InfoSphere
Installation and DataStage Grid Enabled Toolkit in a client’s grid environments, and provides
Deployment integration with Grid Resource Managers, configuration of DataStage,
Workshop QualityStage, and Information Analyzer.
For more details on any of these IBM InfoSphere Professional Services offerings,
and to find a local IBM Information Management contact, visit: the following web
page:
http://www.ibm.com/software/data/services/ii.html

1.4 Workshops for IBM InfoSphere DataStage
Figure 1-4 illustrates the various IBM InfoSphere Services workshops around the
parallel framework of DataStage. The DataFlow Design workshop is part of the
Investigate, Design and Develop track of IBM InfoSphere DataStage workshop.
Information Requirements Investigate,

Data Flow and
Exchange and Definition, Architecture,
Job Design Design,
Discovery and Project Planning
Workshop Workshop
Workshop Develop
Installation Administration, Sizing and

Performance
and Management and Capacity
Tuning
Configuration Production Planning
Workshop
Workshop Automation Workshop Workshop
Installation Sizing and

Performance
and Capacity
Deploy and Configuration Planning
Tuning
Operate Workshop
Workshop Workshop
Figure 1-4 Services workshops for the IBM InfoSphere DataStage Parallel Framework

2
Chapter 2. Data integration overview

Work performed by data integration jobs fall into four general categories:
򐂰 Reading input data, including sequential files, databases and DataStage (DS)
Parallel Datasets
򐂰 Performing row validation to support data quality
򐂰 Performing transformation from data sources to data targets
򐂰 Provisioning data targets

Figure 2-1 depicts the general flow diagram for DataStage parallel jobs.
Before Job Subroutine
Yes
Halt on Error? Exit Failure
No
Create Reject
Files (Limited) Read Input Data
Yes
No
Create Reject Perform
Files (Limited) Validations
Yes
No
Create Reject Perform
Files (Limited) Transformations
Yes
No
Perform Load and/or
Create Reject Create Intermediate Yes
Files (Limited) Over Job Warning Exit Failure
Datasets
Threshold?
No
After Job Subroutine
Figure 2-1 General flow for parallel job

2.1 Job sequences
As shown in Figure 2-1 on page 12, data integration development is intended to
be modular. It is built from individual parallel jobs assembled in IBM InfoSphere
DataStage, controlled as modules from master DataStage Sequence jobs. This
is illustrated in Figure 2-2.
Figure 2-2 Sample job sequence
These job sequences control the interaction and error handling between
individual DataStage jobs, and together form a single end-to-end module in a
DataStage application.
Job sequences also provide the recommended level of integration with external
schedulers (such as AutoSys, Cron, CA7, and so forth). This provides a level of
granularity and control that is easy to manage and maintain, and provides an
appropriate use of the respective technologies.
In most production deployments, job sequences require a level of integration with

various production automation technologies (such as scheduling,
auditing/capture, and error logging).
Chapter 2. Data integration overview 13

2.2 Job types
Nearly all data integration jobs fall into three major types:
򐂰 Transformation jobs
Transformation jobs prepare data for provisioning jobs
򐂰 Provisioning jobs
Provisioning jobs load transformed data, and
򐂰 Hybrid jobs
Hybrid jobs do both.
Table 2-1 defines when each job type is used.
Table 2-1 Job type use

Type Data requirements Example
Transformation Data must not be changed by any Reference tables upon which all subsequent
method unless jobs transforming jobs and the current data target (usually a
an entire subject area have database) depend, or long running provisioning
successfully completed, or where processes. This prevents partial replacement of
the resource requirements for reference data in the event of transformation
data transformation are large. failure, and preserves the compute effort of long
running transformation jobs.
Hybrid Data can be changed regardless Non-reference data or independent data are
of success or failure. candidates. The data target (usually a database)
must allow subsequent processing of error or
reject rows and tolerate partial or complete
non-update of targets. Neither the
transformation nor provisioning requirements
are large.
Provisioning Data must not be changed by any Any target where either all sources have been
method unless jobs transforming successfully transformed or where the resources
an entire subject area have required to transform the data must be
successfully completed, or where preserved in the event of a load failure.
the resource requirements for
data provisioning are large.

2.2.1 Transformation jobs
In transformation jobs, data sources, which might be write-through cache
datasets, are processed to produce a load-ready dataset that represents either
the entire target table or new records to be appended to the target table.
If the entire target table is regenerated with each run, and no other external or
subsequent processing alters the contents of the target table, the output dataset
qualifies as a write-through cache that can be used by subsequent DataStage
jobs instead of reading the entire target table.
When we say “subsequent jobs” we mean any jobs executed as part of the same
transformation cycle, until the target table is updated in the target database by a
provisioning job. A transformation cycle might correspond, for instance, to daily,
weekly, or monthly batch processes.
As a transformation cycle progresses, any real-time updates to the target tables

in the target database (by online or real-time applications, for instance), must
cease, so as not to yield invalid results at the end of the processing cycle.
Otherwise, at the final steps of the processing window, provisioning jobs would
invalidate or overwrite modifications performed by the online applications. The
target table would then become inconsistent.
The following example transformation job demonstrates the use of write-through

cache parallel datasets: The source Oracle stages read data from a couple of
source tables, which are joined and updated by a Parallel Transformer. This
transformed join result is funneled with the content of the target table (extracted
by the “TargetTable_tbl” stage). The result of the Funnel is saved to the Seg1_DS
parallel dataset.
The content of the Seg1_DS parallel dataset can be used by subsequent jobs, in
the same transformation cycle, as input to other joins, lookup, or transform
operations, for instance. Those jobs can thus avoid re-extracting data from that
table unnecessarily. These operations might result in new versions of this
dataset. Those new versions replace Seg1_DS (with different names) and are
used by other subsequent downstream jobs.

An example job is illustrated in Figure 2-3.
Figure 2-3 Sample transformation job with write-through cache

In Figure 2-4, the sample transformation job does not produce write-through
cache. Its sources do not include the target table.
Figure 2-4 Sample transformation job without write-through cache

2.2.2 Hybrid jobs
The sample hybrid job depicted in Figure 2-5 demonstrates a job that transforms
and combines source data, creating a dataset of target rows, and loading the
target table. Because the target table is not sourced, this is an insert of new rows
instead of creating a complete write-through cache.
Figure 2-5 Sample hybrid job

2.2.3 Provisioning jobs
The sample provisioning job depicted in Figure 2-6 demonstrates the
straightforward approach to simple provisioning tasks. In general, it is a good
idea to separate the loading of a database from the ETL process, as a database
load can often fail due to external reasons. Creating the load-ready dataset
allows the provisioning job to be rerun without reprocessing the source extracts
and transformations.
Figure 2-6 Sample provisioning job

3
Chapter 3. Standards
Establishing consistent development standards helps to improve developer
productivity and reduce ongoing maintenance costs. Development standards can
also make it easier to integrate external processes such as automated auditing
and reporting, and to build technical and support documentation.

3.1 Directory structures
IBM InfoSphere DataStage (DS) requires file systems to be available for the
following elements:
򐂰 Software Install Directory
IBM InfoSphere DataStage executables, libraries, and pre-built components.
򐂰 DataStage Project Directory
Runtime information, such as compiled jobs, OSH (Orchestrate® Shell)
scripts, generated BuildOps and Transformers, and logging information.
򐂰 Data Storage
This includes the following storage types:
– DataStage temporary storage: Scratch, temp, buffer
– DataStage parallel dataset segment files
– Staging and archival storage for any source files
By default, these directories (except for file staging) are created during
installation as subdirectories under the base InfoSphere DataStage installation
directory.
3.1.1 Metadata layer

In addition to the file systems listed in the previous section, a DataStage project
also requires a proper amount of space in the Metadata layer (which is a
relational database system). As opposed to the Project directory, the Metadata
layer stores the design time information, including job and sequence designs,
table definitions, and so on.
In this section we do not discuss requirements and recommendations for other

Information Server layers (Metadata and Services). The discussion here is
strictly about the Engine layer.
A single Information Server instance (Services + Metadata) might manage

multiple Engines. The following discussion pertains to the setup of a single
Engine instance.

3.1.2 Data, install, and project directory structures
In Figure 3-1 on page 24 we illustrate how you might configure the file systems to
satisfy the requirements of each class of DataStage storage. These directories
are configured during product installation.
By default, the DataStage Administrator client creates its projects (repositories)

in the Projects directory of the DataStage installation directory. In general, it is a
bad practice to create DataStage projects in the default directory, as disk space
is typically limited in production install directories. For this reason, a separate file
system is created and mounted over the Projects subdirectory.
DataStage parallel configuration files are used to assign resources (such as

processing nodes, disk, and scratch file systems) at runtime when a job is
executed. The DataStage administrator creates parallel configuration files that
define the degree of parallelism (number of nodes, node pools), and resources
used. Parallel configuration files are discussed in detail in the OEM document
Orchestrate 7.5 User Guide, which can be obtained from the IBM CEDI portal,
located at the following web page:
http://www.haifa.ibm.com/ilsl/metadata/cedi.shtml
Data file systems store individual segment files of DataStage parallel datasets.
Scratch file systems are used by the DataStage parallel framework to store
temporary files such as sort and buffer overflow.
Chapter 3. Standards 23
Install File Systems
Install FS Scratch File Systems
/Scratch_<phase>0 /Scratch_<phase>N
/Ascential
/<phase>_Project_A /<phase>_Project_A
/patches
/DataStage /<phase>_Project_Z /<phase>_Project_Z
/DSEngine
Data File Systems

/PXEngine
/Data_<phase>0 /Data_<phase>N
/Configurations
/<phase>_Project_A /<phase>_Project_A
/Projects
1 Gigabyte /<phase>_Project_Z /<phase>_Project_Z
/<phase>_Project_A
Project naming standards include the
deployment phase (dev, it, uat, prod)
prefix, as indicated by <phase>
/<phase>_Project_Z
Figure 3-1 Suggested DataStage install, scratch, and data directories
3.1.3 Extending the DataStage project for external entities

It is suggested that another directory structure, referred to as Project_Plus, be
created to integrate all aspects of a DataStage application that are managed
outside of the DataStage Projects repository. The Project_Plus hierarchy
includes directories for secured parameter files, dataset header files, custom
components, Orchestrate schema, SQL and shell scripts. It might also be useful
to support custom job logs and reports.

The Project_Plus directories provide a complete and separate structure in the
same spirit as a DataStage project, organizing external entities in a structure that
is associated with one corresponding DataStage project. This provides a
convenient vehicle to group and manage resources used by a project.
It is quite common for a DataStage application to be integrated with external

entities, such as the operating system, enterprise schedulers and monitors,
resource managers, other applications, and middle ware. The Project_Plus
directory provides an extensible model that can support this integration through
directories for storing source files, scripts, and other components.
Project_Plus and change management

Project naming conventions recommend naming a project with a prefix to
indicate the deployment phase (dev, it, uat, and prod). Following this naming
convention also separates the associated files in the corresponding Project_Plus
hierarchy.
However, to isolate support files completely and in a manner that is easy to

assign to separate file systems, an additional level of directory structure can be
used to enable multiple phases of application deployment (development,
integration testing, user acceptance testing, and production) as appropriate. If
the file system is not shared across multiple servers, not all of these development
phases might be present on a local file system.
Project_Plus file system

The Project_Plus directory is often stored in the /usr/local home directory (for
example, /usr/local/dstage), but this can be in any location as long as
permissions and file system access are permitted to the DataStage developers
and applications.
Note: The file system where the Project_Plus hierarchy is stored must be
expandable without requiring destruction and re-creation.
Project_Plus directory structure
Figure 3-2 shows typical components and the structure of the Project_Plus
directory hierarchy.
Figure 3-2 Project_Plus directory structure

In Table 3-1 we list the Project_Plus variable descriptions.
Table 3-1 Project_Plus variable descriptions

Directory Description
Project_Plus Top-level of directory hierarchy
/dev Development phase directory tree (if applicable)
/dev_Project_A Subdirectory created for each DataStage project (the actual directory name
“dev_Project_A” should match the corresponding DataStage Project Name).
/bin Location of custom programs, DataStage routines, BuildOps, utilities, and shells
/doc Documentation for programs in /bin subdirectory
/src Source code and Makefiles for items in /bin subdirectory
Depending on change management policies, this directory might only be present

in the /dev development phase directory tree
/datasets Location of parallel dataset header files (.ds files)
/logs Location of custom job logs and reports
/params Location of parameter files for automated program control, a backup copy of
dsenv and backup copies of DSParams:$ProjectName project files
/schemas Location of Orchestrate schema files
/it Integration Test phase directory tree (if applicable)
/uat User Acceptance Test phase directory tree (if applicable)
/prod Production phase directory tree (if applicable)
Project_Plus environment variables
The Project_Plus directory structure is made to be transparent to the DataStage
application, through the use of environment variable parameters used by the
DataStage job developer. Environment variables are a critical portability tool that
enables DataStage applications to be deployed through the life cycle without any
code changes.
In support of a Project_Plus directory structure, the user-defined environment

variable parameters, depicted in Table 3-2, is configured for each project using
the DataStage Administrator, substituting your Project_Plus file system and
project name in the value column:
Table 3-2 Project_Plus environment variables

Name Type Prompt Example Value
PROJECT_PLUS_DATASETS String Project + Dataset /Project_Plus/devProject_A/datasets/

descriptor dir
PROJECT_PLUS_LOGS String Project + Log dir /Project_Plus/devProject_A/logs/
PROJECT_PLUS_PARAMS String Project + /Project_Plus/devProject_A/params/

Parameter file dir
PROJECT_PLUS_SCHEMAS String Project + Schema /Project_Plus/devProject_A/schemas/

dir
PROJECT_PLUS_SCRIPTS String Project + Scripts /Project_Plus/devProject_A/scripts/

dir
Note: the PROJECT_PLUS default values include a trailing directory

separator, to avoid having to specify in the stage properties. This is optional,
but whichever standard the administrator chooses, it is consistently deployed
across projects and job designs.

The Project_Plus environment variables are depicted in Figure 3-3 as they would
appear when loaded into the project dataset.
Figure 3-3 Project_Plus environment variables
In parallel job designs, the Project_Plus parameters are added as job parameters
using the $PROJDEF default value. These parameters are used in the stage
properties to specify the location of DataSet header files, job parameter files,
orchestrate schemas, and external scripts in job flows.
Using Project_Plus with grid or cluster deployments

When deploying a DataStage application in cluster or grid environments, or when
configuring for high availability and disaster recovery scenarios, careful
consideration is made when sharing the Project_Plus file system configuration.
In general, custom components, dataset header files, and other components of

the Project_Plus directory must be visible to all members of the cluster or grid,
using the same mount point on all servers. Creation of small individual mount
points is generally not desirable.
Mount this directory on all members of the cluster after installing IBM InfoSphere
DataStage, but before creating any DataSets.
3.1.4 File staging

Project naming conventions recommend naming a project with a suffix to indicate
the deployment phase (dev, it, uat, prod). Following this naming convention also
separates the associated files in the corresponding Staging hierarchy.
However, to completely isolate support files in a manner that is easy to assign to
separate file systems, an additional level of directory structure can be used to
enable multiple phases of application deployment (development, integration test,
user acceptance test, and production) as appropriate. If the file system is not
shared across multiple servers, not all of these development phases might be
present on a local file system.
In support of the Staging directory structure, the user-defined environment

variable parameters, shown in Table 3-3, is configured for each project using the
DataStage Administrator, substituting your staging file system and project name
in the value column.
Table 3-3 Environment variable parameters

Name Type Prompt Example Value
STAGING_DIR String Staging directory /Staging/
PROJECT_NAME String Project name devProject_A
DEPLOY_PHASE String Deployment phase dev
The Project_Name and Deploy_Phase variables are used to parameterize the

directory location in job designs properly.
Note: The STAGING_DIR default value includes a trailing directory separator,

to avoid having to specify in the stage properties. This is optional, but
whatever standard the administrator chooses, it must be consistently deployed
across projects and job designs.
It is suggested that a separate staging file system and directory structure be

used for storing, managing, and archiving various source data files, as illustrated
in Figure 3-4 on page 31.

Figure 3-4 DataStage staging directories
In each deployment directory, files are separated by project name. See Table 3-4.
Table 3-4 Directory file descriptions
Directory Description
Staging Top-level of directory hierarchy
/dev Development phase directory tree (if applicable)
/dev_Project_A Subdirectory created for each DataStage project (the actual directory name
dev_Project_A should match the corresponding DataStage Project Name)
location of source data files, target data files, error and reject files
/archive Location of compressed archives created by archive process of previously

processed files
/it Integration Test phase directory tree (if applicable)
/uat User Acceptance Test phase directory tree (if applicable)
/prod Production phase directory tree (if applicable)
3.2 Naming conventions
As a graphical development environment, DataStage offers (in certain
restrictions) flexibility to developers when naming various objects and
components used to build a data flow. By default, the Designer tool assigns
default names based on the object type, and the order the item is placed on the
design canvas. Though the default names might create a functional data flow,
they do not facilitate ease of maintenance over time, nor do they adequately
document the business rules or subject areas. Providing a consistent naming
standard is essential to achieve the following results:
򐂰 Maximize the speed of development
򐂰 Minimize the effort and cost of downstream maintenance
򐂰 Enable consistency across multiple teams and projects
򐂰 Facilitate concurrent development
򐂰 Maximize the quality of the developed application
򐂰 Increase the readability of the objects in the visual display medium
򐂰 Increase the understanding of components when seen in external systems
(for example in IBM InfoSphere Business Glossary, Metadata Workbench, or
an XML extract)
In this section we present a set of standards and guidelines to apply to

developing data integration applications using IBM InfoSphere DataStage.
Any set of standards needs to take on the culture of an organization, to be tuned

according to needs, so it is envisaged that these standards shall develop and
adapt over time to suit both the organization and the purpose.
Throughout this section, the term standard refers to those principles that are
required. The term guideline refers to recommended, but not required, principles.
3.2.1 Key attributes of the naming convention

This naming convention is based on the following three-part convention:
Subject, Subject Modifier, and Class Word
In a DataStage job, there might be multiple stages that correlate to a certain area
or task. The naming convention described in the previous section is meant to
reflect that logical organization in a straightforward way. A subject corresponds to
a so-called area in a job, a subject modifier corresponds to a more specific
operation in that area, and the class word indicates the nature of an object.
One example might be a job in which records for data sources such as accounts
and customers are prepared, correlated, credit scores calculated and results

written out as a parallel dataset. In this scenario, as part of the same job, we can
identify a few subjects, such as accounts, customers, and the report generation.
Subject modifiers denote the subset of the logic inside that area. Class word
indicates the type of stage or link.
A few examples, in which subject, subject modifier and class word are separated
by underscores, are shown in Table 3-5.
Table 3-5 Subject examples

Object Name Description
ACCT_FMTCUSTID_TFM A Transformer (TFM) that formats the Customer ID of

account records
RPT_ACCTCUST_LOJN A Left-Outer Join stage (LOJN) that correlates account

and customer records for the report
RPT_SCORE_TFM A Transformer that calculates credit scores
RPT_EXPORT_DS A Dataset stage that exports the report results
In the context of DataStage, the class word is used to identify either a type of
object, or the function that a particular type of object performs. In certain cases
objects can be sub-typed (for example, a Left Outer Join). In these cases the
class word represents the subtype.
For example, in the case of a link object, the class word refers to the functions of
reading, reference (Lookup), moving or writing data (or in a Sequence Job, the
moving of a message).
In the case of a data store the class word refers to the type of data store (as
examples Dataset, Sequential File, Table, View, and so forth).
Where there is no sub classification required, the class word refers to the object.
As an example, a Transformer might be named Data_Block_Split_Tfm.
As a guideline, the class word is represented as a two, three, or four letter

abbreviation. If it is a three or four letter abbreviation, it is word capitalized. If it is
a two letter abbreviation, both letters are capitalized.
A list of frequently-used class word abbreviations is provided in Appendix C,

“DataStage naming reference” on page 391.
One benefit of using the subject, subject modifier, class word approach instead of
using the prefix approach, is to enable two levels of sorting or grouping. In
InfoSphere Metadata Workbench, the object type is defined in a separate field.
There is a field that denotes whether the object is a column, a derivation, a link, a
stage, a job design, and so forth. This is the same or similar information that is
carried in a prefix approach. Carrying this information as a separate attribute
enables the first word of the name to be used as the subject matter, allowing sort
either by subject matter or by object type. In addition, the class word approach
enables sub-classification by object type to provide additional information.
For the purposes of documentation, all word abbreviations are referenced by the
long form to get used to saying the name in full even if reading the abbreviation.
Like a logical name, however, when creating the object, the abbreviated form is
used. This re-enforces wider understanding of the subjects.
The key issue is readability. Though DataStage imposes limitations on the type of
characters and length of various object names, the standard, where possible, is
to separate words by an underscore, which allows clear identification of each
work in a name. This is enhanced by also using word capitalization (for example,
the first letter of each word is capitalized).
3.2.2 Designer object layout

The effective use of naming conventions means that objects need to be spaced
appropriately on the DataStage Designer canvas. For stages with multiple links,
expanding the icon border can significantly improve readability. This approach
takes extra effort at first, so a pattern of work needs to be identified and adopted
to help development. The Snap to Grid feature of Designer can improve
development speed.
When development is more or less complete, attention must be given to the

layout to enhance readability before it is handed over to versioning.
Where possible, consideration must be made to provide DataStage developers

with higher resolution screens, as this provides them with more monitor display
real-estate. This can help make them more productive and makes their work
more easily read.
3.2.3 Documentation and metadata capture

One of the major problems with any development effort, whatever tool you use, is
maintaining documentation. Despite best intentions, and often due to time
constraints, documentation is often something that is left until later or is
inadequately implemented. Establishing a standard method of documentation
with examples and enforcing this as part of the acceptance criteria is strongly
recommended. The use of meaningful naming standards (as outlined in this
section) compliments these efforts.

DataStage provides the ability to document during development with the use of
meaningful naming standards (as outlined in this section). Establishing standards
also eases use of external tools and processes such as InfoSphere Metadata
Workbench, which can provide impact analysis, as well as documentation and
auditing.
3.2.4 Naming conventions by object type

In this section we describe the object type naming conventions.
Projects
Each DataStage Project is a standalone repository. It might have a one-to-one
relationship with an organizations’ project of work. This factor can cause
terminology issues especially in teamwork where both business and developers
are involved.
The name of a DataStage project is limited to a maximum of 18 characters. The

project name can contain alphanumeric characters and underscores.
Projects names must be maintained in unison with source code control. As

projects are promoted through source control, the name of the phase and the
project name should reflect the version, in the following form:
<Phase>_<ProjectName>_<version>
In this example, Phase corresponds to the phase in the application development

life cycle, as depicted in Table 3-6.
Table 3-6 Project phases

Phase Name Phase Description
Dev Development
IT Integration Test
UAT User Acceptance Test
Prod Production
Development Projects
Shared development projects should contain the phase in the life cycle, the
name, and a version number. Examples are shown in the following list:
򐂰 Dev_ProjectA_p0 (initial development phase 0…phase N)
򐂰 Dev_ProjectA_v1 (maintenance)
򐂰 Dev_ProjectA_v2
Individual developers can be given their own sandbox projects, which should
contain the user ID or initials, the application name, the phase in the life cycle
and a version number. This is difficult to do with 18 characters. The following list
shows some examples:
򐂰 JoeDev_ProjectA_v2
򐂰 SueDev_ProjectA_v1
Test projects
Test project names should contain the phase in the life cycle, project name, and
version. The following project names are intended for Integration Testing (IT) and
User Acceptance Testing (UAT):
򐂰 IT_ProjectA_v1_0 (first release)
򐂰 IT_ProjectA_v1_1 (patch or enhancement)
򐂰 UAT_ProjectA_v1_0
Production projects
Although it is preferable to create a new project for each minor and major change
to a production project, making a change to the project name could require
changes to external objects. For example, an enterprise scheduler requires the
project name. Therefore, it is not a requirement that a project name contain
version information.
Using version numbers could allow you to run parallel versions of a DataStage
application, without making changes to the always-on system.
The following list shows examples of acceptable names for the production
project:
򐂰 Prod_ProjectA_v1
򐂰 ProjectA_v1
򐂰 ProjectA
The following examples are project names where the project is single application
focused:
򐂰 Accounting Engine NAB Development is named Dev_AcctEngNAB_v1_0
򐂰 Accounting Engine NAB Production is named Prod_AcctEngNAB
The following examples are project names where the project is multiapplication
focused:
򐂰 Accounting Engine Development or Dev_AcctEngine_v1_0
򐂰 Accounting Engine Production or Prod_AcctEngine

Folder hierarchy
The DataStage repository is organized in a folder hierarchy, allowing related
objects to be grouped together. Folder names can be long, are alpha numeric
and can also contain both spaces and underscores. Therefore, directory names
are word capitalized and separated by either an underscore or a space.
DataStage 7.5 enforced the top level directory structure for various types of
objects, such as jobs, routines, shared containers, and table definitions.
Developers had the flexibility to define their own directory or category hierarchy
beneath that level.
With Information Server 8, objects can be organized in an arbitrary hierarchy of

folders, each folder containing objects of any type. The repository top level is not
restricted to a set of fixed folder names. New top level folders can be added,
such as one per user, subject area, development phase, and so on.
Figure 3-5 presents the top level view, with a list of default folders. As stated
before, objects are not restricted to a top level folder named after the
corresponding type.
Figure 3-5 Default top level folders
Figure 3-6 shows an example of a custom top-level folder that aggregates
objects of several types.
Information Server 8 maintains the restriction that there can only be a single
object of a certain type with a given name.
Figure 3-6 A custom top-level repository folder

Object creation
In Information Server 8, object creation is simplified. To create a new object,
right-click the target parent folder, select New and the option for the desired
object type, as shown in Figure 3-7. Objects of any type can be created in any
folder in the repository hierarchy.
Figure 3-7 Creating a new folder
Categorization by functional module

For a given application or functional module, all objects can be grouped in a
single top-level folder, with sub-levels for separate object types, as in Figure 3-6
on page 38. Job names must be unique in a DataStage project, not only in a
folder.
Categorization by developer
In development projects, folders might be created for each developer as their
personal sandbox. That is the place where they perform unit test activities on
jobs they are developing.
It is the responsibility of each developer to delete unused or obsolete code. The

development manager, to whom is assigned the DataStage Manager role, must
ensure that projects are not inflated with unused objects (such as jobs,
sequences, folders, and table definitions).
Again, object names must be unique in a given project for the given object type.
Two developers cannot save a copy of the same job with the same name in their
individual sandbox categories. A unique job name must be given.
Table definition categories
Unlike DataStage 7.5, in which table definitions were categorized using two level
names (based on the data source type and the data source name), Information
Server 8 allows them to be placed anywhere in the repository hierarchy. This is
depicted in Figure 3-8.
Figure 3-8 A table definition in the repository hierarchy
Figure 3-9 on page 41 shows an example of a table definition stored directly

underneath Table Definitions. Its data source type and data source name
properties do not determine names for parent subfolders.
When saving temporary TableDefs (usually created from output link definitions to
assist with job creation), developers are prompted for the folder in the “Save
Table Definition As” window. The user must pay attention to the folder location,
as these objects are no longer stored in the Table Definition category by default.

Figure 3-9 Table definition categories
Jobs and job sequences

Job names must begin with a letter and can contain letters, numbers, and
underscores only. Because the name can be long, job and job sequence names
must be descriptive and should use word capitalization to make them readable.
Jobs and job sequences are all held under the Category Directory Structure, of
which the top level is the category Jobs.
A job is suffixed with the class word Job and a job sequence is suffixed with the
class word Seq.
The following items are examples of job naming:

򐂰 CodeBlockAggregationJob
򐂰 CodeBlockProcessingSeq
Jobs must be organized under category directories to provide grouping such that
a directory should contain a sequence job and all the jobs that are contained in
that sequence. This is discussed further in “Folder hierarchy” on page 37.
Shared containers
Shared containers have the same naming constraints as jobs in that the name
can be long but cannot contain underscores, so word capitalization must be used
for readability. Shared containers might be placed anywhere in the repository
tree and consideration must be given to a meaningful directory hierarchy. When a
shared container is used, a character code is automatically added to that
instance of its use throughout the project. It is optional as to whether you decide
to change this code to something meaningful.
To differentiate between parallel shared containers and server shared containers,

the following class word naming is recommended:
򐂰 Psc = Parallel Shared Container
򐂰 Ssc = Server Edition Shared Container
Note: Use of Server Shared Containers is discouraged in a parallel job.
Examples of Shared Container naming are as follows:

򐂰 AuditTrailPsc (original naming as seen in the Category Directory)
򐂰 AuditTrailPscC1 (an instance of use of the previously mentioned shared
container)
򐂰 AuditTrailPscC2 (another instance of use of the same shared container)
In the aforementioned examples the characters C1 and the C2 are automatically

applied to the Shared Container stage by DataStage Designer when dragged
onto the design canvas.
Parameters
A parameter can be a long name consisting of alphanumeric characters and
underscores. The parameter name must be made readable using capitalized
words separated by underscores. The class word suffix is parm.
The following examples are of parameter naming:

򐂰 Audit_Trail_Output_Path_parm
򐂰 Note where this is used in a stage property, the parameter name is delimited
by the # sign:
#Audit_Trail_Output_Path_parm#
Links
In a DataStage job, links are objects that represent the flow of data from one
stage to the next. In a job sequence, links represent the flow of a message from
one activity or step to the next.

It is particularly important to establish a consistent naming convention for link
names, instead of using the default DSLink# (where # is an assigned number). In
the graphical Designer environment, stage editors identify links by name. Having
a descriptive link name reduces the chance for errors (for example, during link
ordering). Furthermore, when sharing data with external applications (for
example, through job reporting), establishing standardized link names makes it
easier to understand results and audit counts.
To differentiate link names from stage objects, and to identify in captured

metadata, the prefix lnk_ is used before the subject name of a link.
The following rules can be used to establish a link name:

򐂰 The link name should define the subject of the data that is being moved.
򐂰 For non-stream links, the link name should include the link type (reference,
reject) to reinforce the visual cues of the Designer canvas:
– Ref for reference links (Lookup)
– Rej for reject links (such as Lookup, Merge, Transformer, Sequential File,
and Database)
򐂰 The type of movement might optionally be part of the Class Word. As
examples:
– In for input
– Out for output
– Upd for updates
– Ins for inserts
– Del for deletes
– Get for shared container inputs
– Put for shared container output
򐂰 As data is enriched through stages, the same name might be appropriate for
multiple links. In this case, specify a unique link name in a particular job or job
sequence by including a number. (The DataStage Designer does not require
link names on stages to be unique.)
The following list provides sample link names:
– Input Transactions: lnk_Txn_In
– Reference Account Number Rejects: lnk_Account_Ref_Rej
– Customer File Rejects: lnk_Customer_Rej
Stage names
DataStage assigns default names to stages as they are dragged onto the
Designer canvas. These names are based on the type of stage (Object) and a
unique number, based on the order the object was added to the flow. In a job or
job sequence, stage names must be unique.
Instead of using the full object name, a 2, 3, or 4 character abbreviation must be
used for the class word suffix, after the subject name and subject modifier. A list
of frequently-used stages and their corresponding class word abbreviation can
be found in Appendix C, “DataStage naming reference” on page 391.
Sequencer object naming

In a job Sequencer, links are actually messages. Proceed sequencer links with
the class word msg_ followed by the type of message (as examples, fail and
unconditional), and followed by the ClassName. The following lists shows some
examples:
򐂰 Reception Succeeded Message: msg_ok_Reception
򐂰 Reception Failed Message: msg_fail_Reception
Data stores
For the purposes of this section, a data store is a physical piece of disk storage
where data is held for a period of time. In DataStage terms, this can be either a
table in a database structure or a file contained in a disk directory or catalog
structure. Data held in a database structure is referred to as either a table or a
view. In data warehousing, two additional subclasses of table might be used:
dimension and fact. Data held in a file in a directory structure is classified
according to its type, for example: Sequential File, Parallel Dataset, Lookup File
Set, and so on.
The concepts of “source” and “target” can be applied in a couple of ways. Every
job in a series of jobs could consider the data it gets in to be a source and the
data it writes out as being a target. However, for the sake of this naming
convention a source is only data that is extracted from an original system. A
target is the data structures that are produced or loaded as the final result of a
particular series of jobs. This is based on the purpose of the project: to move
data from a source to a target.
Data stores used as temporary structures to land data between jobs, supporting
restart and modularity, should use the same names in the originating job and any
downstream jobs reading the structure.
Examples of data store naming are as follows:

򐂰 Transaction Header Sequential File or Txn_Header_SF
򐂰 Customer Dimension or Cust_Dim (This optionally can be further qualified as
Cust_Dim_Tgt if you want to qualify it as a final target)
򐂰 Customer Table or Cust_Tab
򐂰 General Ledger Account Number View or GL_Acctno_View

Transformer stage and stage variables
A Transformer stage name can be long (over 50 characters) and can contain
underscores. Therefore, the name can be descriptive and readable through word
capitalization and underscores. DataStage supports two types of Transformers:
򐂰 Tfm: Parallel Transformer
򐂰 BTfm: BASIC Transformer
Note: For maximum performance and scalability, BASIC Transformers

must be avoided in DS parallel data flows.
A Transformer stage variable can have a long name (up to 32 characters)

consisting of alphanumeric characters but not underscores. Therefore, the stage
variable name must be made readable only by using capitalized words. The class
word suffix is stage variable or SV. Stage variables must be named according to
their purpose.
Examples of Transformer stage and stage variable naming are as follows:

򐂰 Alter_Constraints_Tfm
򐂰 recNumSV
When developing Transformer derivation expressions, it is important to

remember stage variable names are case sensitive.
DataStage routines
DataStage BASIC routine names should indicate their function and be grouped
in sub-categories by function under a main category of that corresponds to the
subject area. For example:
Routines/Automation/SetDSParamsFromFile
A how-to document describing the appropriate use of the routine must be

provided by the author of the routine, and placed in a documentation repository.
DataStage custom Transformer routine names should indicate their function and
be grouped in sub-categories by function under a main category that
corresponds to the subject area. For example:
Routines/Automation/DetectTeradataUnicode
Source code, a makefile, and the resulting object for each Custom Transformer
routine must be placed in the Project_Plus source directory. For example:
/Project_Plus/projectA_dev/bin/source
File names
Source file names should include the name of the source database or system
and the source table name or copybook name. The goal is to connect the name
of the file with the name of the storage object on the source system. Source flat
files have a unique serial number composed of the date, “_ETL_” and time. For
example:
Client_Relationship_File1_In_20060104_ETL_184325.psv
Intermediate datasets are created between modules. Their names include the
name of the module that created the dataset or the contents of the dataset in that
more than one module might use the dataset after it is written. For example:
BUSN_RCR_CUST.ds
Target output files include the name of the target subject area or system, the
target table name or copybook name. The goal is the same as with source
files—to connect the name of the file with the name of the file on the target
system. Target flat files have a unique serial number composed of the date,
_ETL_ and time. For example:
Client_Relationship_File1_Out_20060104_ETL_184325.psv
Files and datasets have suffixes that allow easy identification of the content and
type. DataStage proprietary format files have required suffixes. They are
identified in italics in Table 3-7, which defines the types of files and their suffixes.
Table 3-7 File suffixes

File Type Suffix
Flat delimited and non-delimited files .dat.
Flat pipe (|) delimited files .psv
Flat comma-and-quote delimited files .csv.
DataStage datasets .ds.
DataStage filesets .fs
DataStage hash files .hash.
Orchestrate schema files .schema.
Flat delimited or non-delimited REJECT files .rej.
DataStage REJECT datasets _rej.ds.
Flat delimited or non-delimited ERROR files .err.
DataStage ERROR datasets _err.ds.

Flat delimited or non-delimited LOG files .log.
3.3 Documentation and annotation

DataStage Designer provides description fields for each object type. These fields
allow the developer to provide additional descriptions that can be captured and
used by administrators and other developers.
The “Short Description” field is also displayed on summary lines in the Director
and Designer clients. At a minimum, although there is no technical requirement
to do so, job developers should provide descriptive annotations in the “Short
Description” field for each job and job sequence, as in Figure 3-10.
Figure 3-10 Job level short description
In a job, the Annotation tool must be used to highlight steps in a given job flow.
By changing the vertical alignment properties (for example, Bottom) the
annotation can be drawn around the referenced stages, as in Figure 3-11 on
page 48.
Figure 3-11 Sample job annotation
DataStage also allows descriptions to be attached to each stage in the General

tab of the stage properties.
Each stage should have a short description of its function specified in the stage
properties. These descriptions appear in the job documentation automatically
generated from jobs and sequencers adhering to the standards in this document.
More complex operators or operations should have correspondingly longer and
more complex explanations on this tab.
The following list details some examples of such annotations:

򐂰 Job short description
This job takes the data from GBL Oracle Table AD_TYP and does a truncate
load into Teradata Table AD_TYP.
򐂰 ODBC Enterprise stage read

Read the GLO.RcR_GLOBAL_BUSN_CAT_TYP table from
ORACLE_SERVER_parm using the ODBC driver. There are no selection
criteria in the WHERE clause.
򐂰 Oracle Enterprise stage read
Read the GLOBAL.GLOBAL_REST_CHAR table from
ORACLE_SERVER_parm using the Oracle Enterprise operator. There are no
selection criteria in the WHERE clause.
򐂰 Remove Duplicates stage
Removes all but one record with duplicate BUSN_OWN_TYP_ID keys.
򐂰 Lookup stage
– Validates the input and writes rejects
– Validates the input and continues
– Identifies changes and drops records not matched (not updated)
򐂰 Copy stage
– Sends data to the Teradata MultiLoad stage for loading into Teradata, and
to a dataset for use as write-through cache
– Renames and drops columns and is not optimized out
– Is cosmetic and is optimized out
򐂰 Sequential file stage
– Source file for the LANG table
– Target file for business qualification process rejects
򐂰 Transformer stage
– Generates sequence numbers that have a less-than file scope
– Converts null dates
򐂰 Modify stage
Performs data conversions not requiring a Transformer
򐂰 Teradata MultiLoad stage
Loads the RcR_GLOBAL_LCAT_TYP table
򐂰 Dataset stage
– Writes the GLOBAL_Ad_Typ dataset, which is used as write-through
cache to avoid the use of database calls in subsequent jobs.
– Reads the GLOBAL_Lcat dataset, which is used as write-through cache to
avoid the use of database calls.
3.4 Working with source code control systems
The DataStage built-in repository manages objects (jobs, sequences, table
definitions, routines, custom components) during job development. However, this
repository is not capable of managing non-DataStage components (as examples,
UNIX® shell scripts, environment files, and job scheduler configurations) that
might be part of a completed application.
Source code control systems (such as ClearCase®, PVCS and SCCS) are
useful for managing the development life cycle of all components of an
application, organized into specific releases for version control.
As of release 8.1, DataStage does not directly integrate with source code control
systems, but it does offer the ability to exchange information with these systems.
It is the responsibility of the DataStage developer to maintain DataStage objects
in the source code system.
The DataStage Designer client is the primary interface to the DataStage object
repository. Using Designer, you can export objects (such as job designs, table
definitions, custom stage types, and user-defined routines) from the repository as
clear-text format files. These files can be checked into the external source code
control system.
There are three export file format for DataStage 8.X objects:
򐂰 DSX (DataStage eXport format),
򐂰 XML
򐂰 ISX
DSX and XML are established formats that have remained the same since
pre-8.X versions. ISX is a new format introduced in Information Server 8, which
can be imported and exported with the new command-line utility ISTool.
3.4.1 Source code control standards

The first step to effective integration with source code control systems is to
establish standards and rules for managing this process:
򐂰 Establish category naming and organization standard
DataStage objects can be exported individually or by category (folder
hierarchy). Grouping related objects by folder can simplify the process of
exchanging information with the external source code control system. This
object grouping helps establish a manageable middle ground between an
entire project exports and individual object exports.
򐂰 Define rules for exchange with source code control

As a graphical development environment, Designer facilitates iterative job
design. It is cumbersome to require the developer to check-in every change to
a DataStage object in the external source code control system. Rather, rules
must be defined for when this transfer should take place. Typically, milestone
points on the development life cycle are a good point for transferring objects
to the source code control system. For example, when a set of objects has
completed initial development, unit test, and so on.
򐂰 Do not rely on the source code control system for backups
Because the rules defined for transfer to the source code control system are
typically only at milestones in the development cycle, they are not an effective
backup strategy. Furthermore, operating system backups of the project
repository files only establish a point in time, and cannot be used to restore
individual objects.
For these reasons, it is important that an identified individual maintains backup

copies of the important job designs using .DSX file exports to a local or
(preferably) shared file system. These backups can be done on a scheduled
basis by an operations support group, or by the individual DataStage developer.
In either case, the developer should create a local backup prior to implementing
any extensive changes.
3.4.2 Using object categorization standards

As discussed in “Folder hierarchy” on page 37, establishing and following a
consistent naming and categorization standard is essential to the change
management process. The DataStage Designer can export at the project, folder,
and individual object levels. Assigning related objects to the same category
provides a balanced level of granularity when exporting and importing objects
with external source code control systems.
3.4.3 Export to source code control system

The process of exporting DataStage objects to a source code control system is a
straightforward process. It can be done interactively by the developer or project
manager using the Designer client, as explained in this section.
Information Server 8 introduces ISTool, which is a deployment and import/export

tool that can be invoked either on the client or the server side. It extracts and
imports repository objects in a new ISX format.
The major purpose of ISTool is to assist the deployment of DataStage

applications, but this section focuses on its use as an import/export tool.
Client-only tools
The DataStage client includes Windows® command-line utilities for automating
the export process. These utilities (dsexport, dscmdexport, dsimport and
XML2DSX) are documented in the DataStage Designer Client Guide,
LC18-9893.
All exports from the DataStage repository to DSX or XML format are performed
on the Windows workstation.
Exporting with DS Client Tools

The Export menu from DS Designer facilitates the selection of multiple folders
and object types as part of the same export step.
In the example of Figure 3-12, the add link opens up a dialog box from which
individual items or folders can be selected.
Figure 3-12 DS Designer Export facility

The user has already selected the MDMIS R5.1/Parameter Sets category and
one DS routine. Additional items can be selected out of the Select Items window,
and the user can also select the output format of DSX or XML.
Importing with DS client tools

In a similar manner, the import of objects from an external source code control
system is a straightforward process. Import can be interactive through the DS
Designer client or automated through command-line utilities.
In Information Server 8, the DS Designer import interface is basically the same

as the one from DataStage 7.5.
򐂰 Use the source code control system to check-out (or export) the .DSX file to
your client workstation.
򐂰 Import objects in the .DSX file using DS Designer. Choose Import DataStage
Components from the Import menu. Select the file you checked out of your
source code control system by clicking the ellipsis (“…”) next to the filename
field in the import dialog box. This is depicted in Figure 3-13. After selecting
your file, click OK to import.
Figure 3-13 DS Designer import options for DSX files
򐂰 The import of the .DSX file places the object in the same DataStage folder
from which it originated. This means that if necessary it creates the job folder
if it does not already exist.
򐂰 If the objects were not exported with the job executables, then compile the
imported objects from Designer, or from the multi-job compile tool.
There is an equivalent GUI option to import XML files. The import of XML files
first converts the input file from XML to DSX by means of a XSL stylesheet (this
is done behind the scenes). The DSX file is then finally imported into the
repository.
The Windows workstation utilities (dsimport, dscmdimport and XML2DSX) are

documented in the DataStage Designer Client Guide. Job designs in DSX or
XML format can be imported using the Designer client or the dsimport,
dscmdimport or XML2DSX client tools.
Client and server tools

Information Server 8 provides for a new import/export tool named ISTool. It can
be invoked either on the client or the server side. The ability to import and export
objects on the server side is a feature long awaited by the user community.
One can now export an entire project with the following syntax (the same syntax
applies to both the client and server environments):
istool export -domain <domain> -username <user> -password <passwd>
-archive <archive_name> -datastage '<hostname>/project/*.*‘
However, to export just the jobs category the wild card syntax changes a little
(notice the extra /*/). Without this, the folders in are skipped.
istool export -do <domain> -u <user> -p <passwd> -ar <archive_name> -ds
'<hostname>/project/Jobs/*/*.*'
The output or the archive file is a compressed file. If uncompressed, it creates the
directory structure similar to the GUI. The import option can be used to import
the .isx (archive suffix), similar to the .xml or .dsx files.
It is much faster to run the export/import on the server side with the ISTool when
compared to the DS Designer client tools described in 3.4.3, “Export to source
code control system” on page 51.
The ISTool is documented in the IBM Information Server Manager User Guide,
LC19-2453-00.

4
Chapter 4. Job parameter and

environment variable
management
An overview of DataStage (DS) job parameters can be found in the IBM
InfoSphere DataStage and QualityStage Designer Client Guide, LC18-9893.
DataStage jobs can be parameterized to allow for portability and flexibility.

Parameters are used to pass values for variables into jobs at run time. There are
two types of parameters supported by DataStage jobs:
򐂰 Standard job parameters
– Are defined on a per job basis in the job properties dialog box.
– The scope of a parameter is restricted to the job.
– Used to vary values for stage properties, and arguments to before/after job
routines at runtime.
– No external dependencies, as all parameter metadata is a sub element to
a single job.

򐂰 Environment variable parameters:
– Use operating system environment variable concept.
– Provide a mechanism for passing the value of an environment variable into
a job as a job parameter (Environment variables defined as job
parameters start with a $ sign).
– Are similar to a standard job parameter in that it can be used to vary
values for stage properties, and arguments to before/after job routines.
– Provide a mechanism to set the value of an environment variable at
runtime. DataStage provides a number of environment variables to enable
/ disable product features, fine-tune performance, and to specify runtime
and design time functionality (for example, $APT_CONFIG_FILE).

4.1 DataStage environment variables
In this section we discuss the DataStage environment variables, including scope,
special values, an overview of predefined variables and recommended settings
for all jobs, and the migration of project-level settings across projects.
4.1.1 DataStage environment variable scope

Although operating system environment variables can be set in multiple places,
there is a defined order of precedence that is evaluated when a job’s actual
environment is established at runtime. The scope of an environment variable is
dependent upon where it is defined. Table 4-1 shows where environment
variables are set and order of precedence.
Table 4-1 Environment variables

Where Defined Scope (* indicates highest precedence)
System profile System wide
dsenv All DataStage processes
Shell script (if dsjob –local Only DataStage processes spawned by dsjob
is specified)
Project All DataStage processes for a project
Job Sequencer Job sequence and sequence sub processes
Job * Current job’s environment and sub processes
The daemon for managing client connections to the DataStage engine is called
dsrpcd. By default (in a root installation), dsrpcd is started when the server is
installed, and should start whenever the machine is restarted. dsrpcd can also be
manually started and stopped using the $DSHOME/uv –admin command. For more
information, see the IBM InfoSphere DataStage and QualityStage Administrator
Client Guide, LC18-9895.
By default, DataStage jobs inherit the dsrpcd environment which, on UNIX

platforms, is set in the etc/profile and $DSHOME/dsenv scripts. On Windows, the
default DataStage environment is defined in the registry. Client connections do
not pick up per-user environment settings from their $HOME/.profile script.
Environment variable settings for particular projects can be set in the DataStage
Administrator client. Any project-level settings for a specific environment variable
override any settings inherited from dsrpcd.
Chapter 4. Job parameter and environment variable management 57

In DataStage Designer, environment variables might be defined for a particular
job using the Job Properties dialog box. Any job-level settings for a specific
environment variable override any settings inherited from dsrpcd or from
project-level defaults. Project-level environment variables are set and defined in
DataStage Administrator.
4.1.2 Special values for DataStage environment variables

To avoid hard-coding default values for job parameters, there are three special
values that can be used for environment variables in job parameters. They are
described in Table 4-2.
Table 4-2 Environment variable special values

Value Use
$ENV Causes the value of the named environment variable to be retrieved

from the operating system of the job environment. Typically this is used
to pickup values set in the operating system outside of DataStage.
$PROJDEF Causes the project default value for the environment variable (as
shown on the Administrator client) to be picked up and used to set the
environment variable and job parameter for the job.
$UNSET Causes the environment variable to be removed completely from the

runtime environment. Several environment variables are evaluated
only for their presence in the environment (for example,
APT_SORT_INSERTION_CHECK_ONLY).
Note: $ENV must not be used for specifying the default $APT_CONFIG_FILE
value because, during job development, Designer parses the corresponding
parallel configuration file to obtain a list of node maps and constraints
(Advanced stage properties)
4.1.3 Environment variable settings

In this section we call discuss a number of key environment variables, and
provide descriptions of their settings for parallel jobs.

Environment variable settings for all jobs
IBM suggests the following environment variable settings for all DataStage
parallel jobs. These settings can be made at the project level, or set on an
individual basis in the properties for each job. It might be helpful to create a
parameter set that includes the environment variables described in Table 4-3.
Table 4-3 Environment variable descriptions

Environment Variable Setting Description
$APT_CONFIG_FILE filepath Specifies the full path name to the parallel configuration file.
This variable must be included in all job parameters so that it
can be easily changed at runtime.
$APT_DUMP_SCORE Outputs parallel score dump to the DataStage job log,

providing detailed information about actual job flow including
operators, processes, and Datasets. Extremely useful for
understanding how a job actually ran in the environment. See
Appendix E, “Understanding the parallel job score” on
page 401.
$OSH_ECHO Includes a copy of the generated osh (Orchestrate Shell) in the

job’s DataStage log.
$APT_RECORD_ Outputs record counts to the DataStage job log as each

COUNTS operator completes processing. The count is per operator per
partition.
This setting must be disabled by default, but part of every job
design so that it can be easily enabled for debugging purposes.
$APT_PERFORMANCE $UNSET If set, specifies the directory to capture advanced job runtime
_DATA performance statistics.
$OSH_PRINT_ Outputs actual runtime metadata (schema) to DataStage job

SCHEMAS log.
This setting must be disabled by default, but part of every job
design so that it can be easily enabled for debugging purposes.
$APT_PM_SHOW_ Places entries in DataStage job log showing UNIX process ID

PIDS (PID) for each process started by a job. Does not report PIDs
of DataStage “phantom” processes started by Server shared
containers.
$APT_BUFFER_ Maximum buffer delay in seconds

MAXIMUM_TIMEOUT
On Solaris platforms only: When working with large parallel datasets (where
the individual data segment files are larger than 2 GB), you must define the
environment variable $APT_IO_NOMAP.

On Tru64 5.1A platforms only: On Tru64 platforms, the environment variable
$APT_PM_NO_SHARED_MEMORY must be set to 1 to work around a
performance issue with shared memory MMAP operations. This setting
instructs the parallel framework to use named pipes rather than shared
memory for local data transport.
Additional environment variable settings

Throughout this document, a number of environment variables are mentioned for
tuning the performance of a particular job flow, assisting in debugging, or
changing the default behavior of specific DS Parallel stages. The environment
variables mentioned in this document are summarized in Appendix G,
“Environment variables reference” on page 413. An extensive list of environment
variables is documented in the IBM InfoSphere DataStage and QualityStage
Parallel Job Advanced Developer Guide, LC18-9892.
4.1.4 Migrating project-level environment variables

When migrating projects between machines or environments, it is important to
note that project-level environment variable settings are not exported when a
project is exported. These settings are stored in a file named DSPARAMS in the
project directory. If an environment variable has not been configured for the
project, the migrated job fails during startup.
Any project-level environment variables must be set for new projects using the
Administrator client, or by carefully editing the DSPARAMS file in the project.
Note: Copying DSPARAMS between versions of DataStage might result in

failure.
In addition, migrating lists of environment variable parameters and their values

from one project to another can be managed by the administrator using a CEDI
developed tool called Environment Variable Transporter.
4.2 DataStage job parameters

Parameters are passed to a job as either DataStage job parameters or as
environment variables. The naming standard for job parameters uses the suffix
_parm in the variable name. Environment variables have a prefix of $ when used
as job parameters.

Job parameters are passed from a job sequencer to the jobs in its control as
though a user were answering the runtime dialog questions displayed in the
DataStage Director job-run dialog. Job parameters can also be specified using a
parameter set.
The scope of job parameters depends on their type. The scope depends on the
following factors:
򐂰 Is specific to the job in which it is defined and used. Job parameters are
stored internally in DataStage for the duration of the job, and are not
accessible outside that job.
򐂰 Can be extended by the use of a job sequencer, which can manage and pass
the job parameter among jobs in the sequence.
4.2.1 When to use parameters

As a standard practice, file names, database names, passwords, and message
queue names must be parameterized. It is left to the discretion of the developer
to parameterize other properties. When deciding on what to parameterize, ask
the following questions:
򐂰 Could this stage or link property be required to change from one project to
another?
򐂰 Could this stage or link property change from one job execution to another?
If you answer yes to either of these questions, you should create a job parameter
and set the property to that parameter.
To facilitate production automation and file management, the Project_Plus and

Staging file paths are defined using a number of environment variables that must
be used as job parameters. (See 3.1.3, “Extending the DataStage project for
external entities” on page 24 and 3.1.4, “File staging” on page 29.)
Job parameters are required for the following DataStage programming elements:
򐂰 File name entries in stages that use files or datasets must never use a
hard-coded operating system path name.
– Staging area files must always have path names as follows:
/#$STAGING_DIR##$DEPLOY_PHASE_parm#[filename.suffix]
– DataStage datasets always have path names as follows:
/#$PROJECT_PLUS_DATASETS#[headerfilename.ds]
򐂰 Database stages must always use variables for the server name, schema (if
appropriate), user ID and password.
A list of recommended job parameters is summarized in Appendix D, “Example
job template” on page 397.

4.2.2 Parameter standard practices
File Name stage properties must be configured using two parameters, one for
directory path and the second for file name. The directory path delimiter must be
specified in the property to avoid errors. Do not assume the runtime value of the
directory parameter includes the appropriate delimiter. If the user supplies it the
operating system accepts // as a delimiter, and if it is not provided, which is
common, the file name property specification is correct.
Example of standard practice for file name properties:

#Dir_Path#/#File_Name#
Similar to directory path delimiters, database schema names, etc. should contain
any required delimiter.
Example of standard practice for table name properties:

#DatabaseSchemaName#.TableName
User Accounts and passwords must be specified as environment variables.
Passwords must be set to type encrypted, and the default value maintained using
the DataStage Administrator.
4.2.3 Specifying default parameter values

Default values of job parameters migrate with the job. If the value is not
overridden then you run the risk of unintentionally using the wrong resources,
such as connecting to the wrong database or referencing the wrong file. To
mitigate this risk developers should follow this standard practice when setting the
default value of a parallel job’s parameter:
򐂰 Standard parameters: Ensure the default value is empty
򐂰 Environment variable parameters: Ensure the default value is empty, $ENV,
$PROJDEF or $UNSET
The intent of this standard practice is to ensure a job is portable. It thus requires
the value of a parameter to be set independent of the job. During development of
a job, consider using the standard practice of always using a test harness
sequencer to execute a parallel job. The test harness allows the job to be run
independently and ensures the parameter values are set. When the job is ready
for integration into a production sequencer, the test harness can be cut, pasted,
and linked into the production sequencer. The test harness is also useful in test
environments, as it allows you to run isolated tests on a job.

In rare cases, normal parameter values must be allowed default values. For
example, a job might be configured with parameters for array_size or
commit_size and the corresponding Database stage properties set to these
parameters. The default value must be set to an optimal value as determined by
performance testing. This value must be relatively static. The value can always
be overridden by job control. This exception also minimizes the number of
parameters. Consider that array_size is different for every job. You can have a
unique parameter for every job, but it is difficult to manage the values for every
job.
4.2.4 Parameter sets

One of the great new features in Information Server 8 is the concept of
parameter sets. They greatly simplify the management of job parameters by
providing the following elements:
򐂰 Central location for the management of large lists of parameters
򐂰 Simplification of the propagation of parameter sets into jobs, sequences and
shared containers
򐂰 Convenient way of defining multiple sets of values in the form of value files
Parameter sets are such an advance compared to earlier versions of DataStage

that they actually render obsolete the techniques previously devised and
recommended (such as Basic routines to read values from files and the use of
job templates or shared containers as a way of incorporating predefined sets of
parameters into new jobs).
A parameter set is assigned a name, and as such can be passed into jobs,
shared containers and sequences collectively. It is an entity on its own, stored
anywhere in the repository tree. We recommend creating a folder named
“parameter sets” for this purpose.

Figure 4-1 shows an example of a parameter set named MDMIS. The
Parameters tab lists the individual parameter names and their default values,
types, prompts and descriptions.
Figure 4-1 A sample parameter set
The multiple values that parameter set MDMIS can assume are defined in the
Values tab. The list of parameters is presented horizontally. The first column is
the name of a value file. Subsequent columns contain values for each individual
parameter.
In this example, there is a single value file, but there might be multiple such value
files for the same parameter set. This is depicted in Figure 4-2 on page 65.
Parameter sets are stored in the metadata layer along with the rest of the
project’s design metadata. They might be exported and imported individually or
with other objects using the DS Designer’s Import/Export facilities.

Figure 4-2 Value files of a parameter set
Value files are actual flat files stored in the DS Project in the DSEngine host file
system. Figure 4-2 presents an example of a shell session displaying the location
and content of the STATIC_MDMIS value file for the MDMIS parameter set.
There is a directory named ParameterSets/MDMIS under the MDMRDP project
directory. The value file STATIC_MDMIS is stored in that directory. The output
showing the first 10 lines of output is depicted in Figure 4-3.
Figure 4-3 Location and content of a value file in the DS project directory

Propagating parameter sets
Propagating large sets of parameters is as easy as defining a single parameter.
Instead of repeatedly defining individual parameters in each sequence, job, and
shared container, all it takes is defining a single parameter of type Parameter Set
(click the Add Parameter Set button).
Figure 4-4 shows a sample job sequence, with a parameter set named MDMIS.
Sequence
Figure 4-4 Job sequence with a MDMIS parameter set

Figure 4-5 displays how a sequence passes the parameter set down to one of its
job activities. The elimination of clutter is evident.
The parameters for the job invoked by this job activity are defined in a way similar
to the one depicted in Figure 4-4 on page 66.
Sequence
Figure 4-5 Passing a parameter set to a job activity

Setting the runtime value for a parameter set
Upon job or sequence invocation, the value file for a parameter set might be
specified on the director job run options, as depicted in Figure 4-6.
Figure 4-6 Setting the value file upon job sequence invocation
Values for individual parameters might be overridden manually by the operator, in

addition to the ones assumed by the choice of value file. For dsjob, the command
line would follow a syntax similar to setting an individual parameter. For the
example, the choice of value file is specified as:
param MDMIS=STATIC_MDMIS
Using parameter sets is strongly encouraged in all types of DataStage projects.
Environment variable parameter lists

When a job is migrated from one project to another, the job fails during startup if
the environment variable has not been configured for the project. Migrating lists
of environment variable parameters and their values from one project to another
can be managed by the administrator using a CEDI developed tool called the
Environment Variable Transporter.
If the tool is not available then you must enter the environment variable
parameters one at a time in the DataStage Administrator. If this becomes too
cumbersome, consider the fact that environment variable parameters are stored
in the DSParams file. The DSParams is a text file that can be modified by hand.
However, if you choose to modify this file by hand you do so at your own risk.

5
Chapter 5. Development guidelines

In this chapter we discuss and describe guidelines for developing DataStage
(DS) parallel jobs. The guidelines in this chapter relate to the following subjects:
򐂰 Modular development
򐂰 Establishing job boundaries
򐂰 The use of job design templates
򐂰 Parallel shared containers
򐂰 Error and reject record handling
򐂰 Considerations for usage and impact analysis
In addition, we also include a discussion on the usage of components, such as

Server Edition components, Copy stage, Parallel Datasets, Parallel Transformer
and BuildOps.
We then delve into a greater level of detail on how to use the various stage types
in subsequent chapters of this book.

5.1 Modular development
Modular development techniques must be used to maximize re-use of DataStage
jobs and components:
򐂰 Job parameterization allows a single job design to process similar logic
instead of creating multiple copies of the same job. The multiple-instance job
property allows multiple invocations of the same job to run simultaneously.
򐂰 A set of standard job parameters must be used in DataStage jobs for source
and target database parameters (such as DSN and user, password) and
directories where files are stored. To ease re-use, these standard parameters
and settings must be made part of a Designer job template.
򐂰 Create a standard directory structure outside of the DataStage project
directory for source and target files, intermediate work files, and so forth.
򐂰 Where possible, create reusable components such as parallel shared
containers to encapsulate frequently used logic.
5.2 Establishing job boundaries

It is important to establish appropriate job boundaries when developing parallel
jobs. In certain cases, functional requirements might dictate job boundaries. For
example, it might be appropriate to update all dimension values before inserting
new entries in a data warehousing fact table. But functional requirements might
not be the only factor driving the size of a given DataStage job.
Though it might be possible to construct a large, complex job that satisfies given
functional requirements, this might not be appropriate. The following list details
factors to consider when establishing job boundaries:
򐂰 Establishing job boundaries through intermediate datasets creates
checkpoints that can be used in the event of a failure when processing must
be restarted. Without these checkpoints, processing must be restarted from
the beginning of the job flow. It is for these reasons that long-running tasks
are often segmented into separate jobs in an overall sequence.
– For example, if the extract of source data takes a long time (such as an
FTP transfer over a wide area network) land the extracted source data to a
parallel data set before processing. To continue processing to a database
can cause conflict with locking tables when waiting for the FTP to
complete.
– As another example, land data to a parallel dataset before loading to a
target database unless the data volume is small, the overall time to
process the data is minimal, or if the data volume is so large that it cannot
be staged on the extract, transform, and load (ETL) server.

򐂰 Larger, more complex jobs require more system resources (CPU, memory,
swap) than a series of smaller jobs, sequenced together through intermediate
datasets. Resource requirements are further increased when running with a
greater degree of parallelism specified by a given configuration file. However,
the sequence of smaller jobs generally requires more disk space to hold
intermediate data, and the speed of the I/O subsystem can impact overall
end-to-end throughput.
In 12.3, “Minimizing runtime processes and resource requirements” on
page 179, we provide recommendations for minimizing resource
requirements of a given job design, especially when the volume of data does
not dictate parallel processing.
򐂰 Breaking large job flows into smaller jobs might further facilitate modular
development and re-use if business requirements for more than one process
depend on intermediate data created by an earlier job.
򐂰 The size of a job directly impacts the speed of development tasks such as
opening, saving, and compiling. These factors might be amplified when
developing across a wide-area or high-latency network connection. In
extreme circumstances, this can significantly impact developer productivity
and ongoing maintenance costs.
򐂰 The startup time of a given job is directly related to the number of stages and
links in the job flow. Larger more complex jobs require more time to startup
before actual data processing can begin. Job startup time is further impacted
by the degree of parallelism specified by the parallel configuration file.
򐂰 Remember that the number of stages in a parallel job includes the number of
stages in each shared container used in a particular job flow.
As a rule of thumb, keeping job designs to less than 50 stages is a good starting
point. But this is not a hard-and-fast rule. The proper job boundaries are
ultimately dictated by functional/restart/performance requirements, expected
throughput and data volumes, degree of parallelism, number of simultaneous
jobs and their corresponding complexity, and the capacity and capabilities of the
target hardware environment.
Combining or splitting jobs is relatively easy, so do not be afraid to experiment

and see what works best for your jobs in your environment.
Chapter 5. Development guidelines 71

5.3 Job design templates
DataStage Designer provides the developer with re-usable job templates, which
can be created from an existing parallel job or job sequence using the New
Template from Job command.
Template jobs must be created with the following elements:

򐂰 Standard parameters (for example, source and target file paths, database
login properties, and so forth)
򐂰 Environment variables and their default settings
򐂰 Project_Plus environment variables (see 3.1.3, “Extending the DataStage
project for external entities” on page 24.)
򐂰 Annotation blocks
In addition, template jobs might contain any number of stages and pre-built logic,
allowing multiple templates to be created for various types of standardized
processing.
5.4 Default job design

Default job designs include all of the capabilities detailed in Chapter 3,
“Standards” on page 21. Template jobs should contain all the default
characteristics and parameters the project requires. These defaults provide at a
minimum:
򐂰 Development phase neutral storage (as examples: dev, it, uat and prod)
򐂰 Support for Teradata, Oracle, DB2/UDB and SQL Server login requirements
򐂰 Enforced project standards
򐂰 Optional operational metadata (runtime statistics) suitable for loading into a
database
򐂰 Optional auditing capabilities
The default job design specifically supports the creation of write-through cache in
which data in load-ready format is stored in parallel datasets for use in the load
process or in the event the target table becomes unavailable.
The default job design incorporates several features and components of

DataStage that are used together to support tactical and strategic job
deployment.

These features include:
򐂰 Restartable job sequencers that manage one or more jobs, detect and report
failure conditions, provide monitoring and alert capabilities, and support
checkpoint restart functionality.
򐂰 Custom routines written in DataStage BASIC (DS Basic) that detect external
events, manage and manipulate external resources, provide enhanced
notification and alert capabilities, and interface to the UNIX operating system
򐂰 DataStage Parallel jobs that exploit job parameterization, runtime UNIX
environment variables, and conditional execution.
Each subject area is broken into sub-areas and each sub-area might be further
subdivided. These sub-areas are populated by a DataStage job sequencer using
two types of DataStage jobs at a minimum:
򐂰 A job that reads source data and then perform one of the following tasks
– Transforms it to load-ready format
– Optionally stores its results in a write-through cache DataStage dataset or
loads the data to the target table.
򐂰 A job that reads the DataStage dataset and loads it to the target table.
Other sections discuss in detail each of the components and give examples of
their use in a working example job sequencer.
5.5 Parallel shared containers

Parallel shared containers allow common logic to be shared across multiple jobs.
For maximum component re-use, enable Runtime Column Propagation (RCP) at
the project level and for every stage in the parallel shared container. This allows
the container input and output links to contain only the columns relevant to the
container processing. When using RCP, any additional columns are passed
through the container at runtime without the need to separate and remerge.
Because parallel shared containers are inserted when a job is compiled, all jobs
that use a shared container must be recompiled when the container is changed.
The Usage Analysis and Multi-Job Compile tools can be used to recompile jobs
that use a shared container.

5.6 Error and reject record handling
Reject rows are those rows that fail active or passive business rule driven
validation as specified in the job design document. Error rows are those rows
caused by unforeseen data events such as values too large for a column or text
in an unsupported language.
The exact policy for each reject is specified in the job design document, and
further, whether the job or ETL processing is to continue is specified on a per-job
and per-sequence and per-script basis based on business requirements.
Reject files include those records rejected from the ETL stream due to
Referential Integrity failures, data rule violations or other reasons that would
disqualify a row from processing. The presence of rejects might indicate that a
job has failed and prevent further processing. Specification of this action is the
responsibility of the Business Analyst and is published in the design document.
Error files include those records from sources that fail quality tests. The presence
of errors might not prevent further processing. Specification of this action is the
responsibility of the Business Analyst and is published in the design document.
Both rejects and errors are archived and placed in a special directory for
evaluation or other action by support staff. The presence of rejects and errors are
detected and notification sent by email to selected staff. These activities are the
responsibility of job sequencers used to group jobs by reasonable grain or by a
federated scheduler.
ETL actions to be taken for each record type is specified for each stage in the job
design document. These actions include:
1. Ignore – some process or event downstream of the ETL process is
responsible for handling the error.
2. Reprocess – rows are reprocessed and re-enter the data stream.
3. Push back – rows are sent to a Data Steward for corrective action.
The default action is to push back reject and error rows to a Data Steward.

5.6.1 Reject handling with the Sequential File stage
The Sequential File stage can optionally include a reject link, which outputs rows
that do not match the given table definition and format specifications. By default,
rows that cannot be read are dropped by the Sequential File stage. A message is
always written to the Director log which details the count of rows successfully
read and rows rejected.
The Sequential File stage offers the reject options listed in Table 5-1:
Table 5-1 Reject options

Option Description
Continue Drop read failures from input stream. Pass successful reads to the output
stream. (No reject link exists)
Fail Abort job on read format failure (No reject link exists)
Output Reject switch failures to the reject stream. Pass successful reads to the
output stream. (Reject link exists)
The reject option must be used in all cases where active management of the
rejects is required.
If a file is created by this option, it must have a *.rej file extension. Alternatively, a
shared container error handler can be used.
Rejects are categorized in the ETL job design document using the ranking listed
in Table 5-2.
Table 5-2 Reject ranking

Category Description Sequential File stage Option
Rejects are expected and can Use the Continue option. Only records
be ignored that match the given table definition
and format are output. Rejects are
tracked by count only.
Rejects should not exist but Use the Output option. Send the reject
should not stop the job, and stream to a *.rej file.
must be reviewed by the Data
Steward.
Rejects should not exist and Use the Fail option.

should stop the job.

5.6.2 Reject handling with the Lookup stage
The Lookup stage compares a single input stream to one or more reference
streams using keys, and rejects can occur if the key fields are not found in the
reference data. This behavior makes the Lookup stage valuable for positive
(reference is found) and negative (reference is not found) business rule
validation. DataStage offers options in a Parallel Lookup stage, as listed in
Table 5-3.
Table 5-3 Parallel Lookup stage options

Option Description
Continue Ignore Lookup failures and pass Lookup fields as nulls to the output
stream. Pass successful Lookups to the output stream.
Drop Drop Lookup failures from the input stream. Pass successful Lookups
to the output stream.
Fail Abort job on Lookup failure
Reject Reject Lookup failures to the reject stream. Pass successful Lookups
to the output stream.
rejects is required. Furthermore, to enforce error management only one
reference link is allowed on a Lookup stage. If there are multiple validations to
perform, each must be done in its own Lookup.
If a file is created by this option, it must have a *.rej or *.err file extension. The
*.rej extension is used when rejects require investigation after a job run, the *.err
extension when rejects can be ignored but need to be recorded. Alternatively, a
local error handler based on a shared container can be used.

Rejects are categorized in the ETL job design document using the ranking in
Table 5-4.
Table 5-4 Reject ranking: Lookup stage

Category Description Lookup stage option
Rejects are expected and Drop if lookup fields are necessary down
can be ignored stream or Continue if lookup fields are
optional
Rejects can exist in the data, Send the reject stream to an *.err file or
however, they only need to tag and merge with the output stream.
be recorded but not acted on.
Rejects should not exist but Send the reject stream to an *.rej file or
should not stop the job, and tag and merge with the output stream.
must be reviewed by the
Data Steward.
Rejects should not exist and Use the Fail option.

should stop the job.
5.6.3 Reject handling with the Transformer stage

Rejects occur when a Transformer stage is used and a row:
1. satisfies requirements for a reject conditional output stream.
2. cannot satisfy requirements of any conditional output stream and is rejected
by the default output stream.
If a file is created from the reject stream, it must have a *.rej or *.err file extension.
The *.rej extension is used when rejects require investigation after a job run, the
*.err extension when rejects can be ignored but need to be recorded.
Alternatively, a shared container error handler can be used.

in Table 5-5.
Table 5-5 Reject ranking: Transformer stage

Category Description Transformer stage option
Rejects are expected and can be Funnel the reject stream back to the
ignored. output streams.
Rejects can exist in the data, Send the reject stream to an *.err file
however, they only need to be or tag and merge with the output
recorded but not acted on. stream.
Rejects should not exist but Send the reject stream to an *.rej file
should not stop the job, and be or tag and merge with the output
reviewed by the Data Steward. stream.
Rejects should not exist and Send the reject stream to a reject file
should stop the job. and halt the job.
5.6.4 Reject handling with Target Database stages

Database stages (such as DB2/UDB Enterprise, ODBC Enterprise, and Oracle
Enterprise) offer an optional reject link that can be used to capture rows that
cannot be written to the target database. To capture rejects from a target
database, a reject link must exist on that stage. Otherwise, reject rows are not
captured. A message is always written to the Director log which details the count
of rows successfully read and rows rejected.
Target Database stages offer the reject options listed in Table 5-6.
Table 5-6 Reject options

Option Description
No reject link exists Do not capture rows that fail to be written.
Reject link exists Pass rows that fail to be written to the reject stream.
rejects is required.
If a file is created by this option, it must have a *.rej file extension. Alternatively, a
shared container error handler is used.

in Table 5-7.
Table 5-7 Reject ranking

Category Description Target Database stage Option
Rejects are expected and can be No reject link exists. Only records
ignored that match the given table definition
and database constraints are
written. Rejects are tracked by count
only.
Rejects should not exist but should Reject link exists. Send the reject
not stop the job, and must be stream to a *.rej file.
reviewed by the Data Steward.
5.6.5 Error processing requirements

Jobs produce flat files containing rejects and errors. They might alternatively
process rows on reject ports and merge these rows with the normal output
stream. This section deals with both methods of handling errors.
Processing errors and rejects to a flat file

Each job produces a flat file for errors and a flat file for rejects with a specific
naming convention:
1. The project name ($PROJECT_NAME) and a underscore “_”;
2. The job name (JOB_NAME_parm) and a underscore “_”;
3. The project phase ($DEPLOY_PHASE) and a underscore “_”;
4. The job run identifier (RUN_ID_parm) and a period “.”; and
5. The appropriate file type, one of “rej” or “err”.
For example, job DECRP_N_XformClients in the ECR_FACTS project in the

development environment with a run identifier of 20060201-ETL-091504 would
have the following reject and error file names:
򐂰 ECR_FACTS_DECRP_N_XformClients_dev_20060201-ETL-091504.rej
򐂰 ECR_FACTS_DECRP_N_XformClients_dev_20060201-ETL-091504.err
Rows are converted to the common file record format with 9 columns (as shown
in Figure 5-2 on page 81) using Column Export and Transformer stages for each
reject port, and gathered using a Funnel stage that feeds a Sequential File stage.
The Column Export and Transformer stages might be kept in a template Shared
Container the developer makes local in each job.

The standard columns for error and reject processing are listed in Table 5-8.
Table 5-8 Standard columns for error and reject processing

Column Name Key? Data Source
HOST_NAME Yes DSHostName Transformer macro in the error handler
PROJECT_NAME Yes DSProjectName Transformer macro in the error handler
JOB_NAME Yes DSJobName Transformer macro in the error handler
STAGE_NAME Yes The name of the stage from which the error came
DATA_OBJ_NAME Yes The source table or file data object name
RUN_ID Yes RUN_ID_parm
ETL_ROW_NUM Yes Data stream coming in to the error handler
ROW_DATA No The columns from the upstream stages reject port

exported to a single pipe-delimited “|” varchar column
using the Column Export stage in the error handler
Length to be determined by max length of record to be
maintained as row data.
In Figure 5-1 we depict the stages that process the errors produced by a job.
The Column Export

Stage maps the unique
columns to the single
standard column.
The Transformer Stage

adds the required key
columns.
Figure 5-1 Error processing components

The input to the Column Export stage explicitly converts the data unique to the
reject stream (in this case, Track*) to a single output column, ROW_DATA, as
Figure 5-2 Error processing Column Export stage

And the downstream Transformer stage builds the standard output record by
creating the required keys, as depicted in Figure 5-3.
Figure 5-3 Error processing Transformer stage

Processing errors and rejects and merging with an output
stream
There might be processing requirements specifying that rejected or error rows be
tagged as having failed a validation and be merged back into the output stream.
This is done by processing the rows from the reject ports and setting the value of
a specific column with a value specified by the design document. In Table 5-9 we
identify the tagging method to be used for the previously cited operators.
Table 5-9 Tagging method

stage Description Method
Lookup A failed Lookup rejects an intact input Connect the reject port to a Transformer stage
row whose key fails to match the where those columns selected for replacement
reference link key. One or more are set to specific values. Connect the output
columns might have been selected for stream of the Transformer and Lookup stages
replacement when a reference key is to a Funnel stage to merge the two streams.
found.
Switch A failed switch rejects an intact input Connect the reject port to a Transformer stage
row show key fails to resolve to one of where columns are set to specific values.
the switch output stream. Connect the output stream of the Transformer
Stage and one or more output streams of the
Switch stage to a Funnel stage to merge the
two (or more) streams.
Transformer A Transformer rejects an intact input Connect the reject port to a Transformer stage
row that cannot pass conditions where columns are set to specific values.
specified on the output streams, OR Connect the output stream of the corrective
with columns contain illegal values for Transformer stage and one or more output
operations performed on said streams of the original Transformer stage to a
columns. In either case, attaching a Funnel stage to merge the two (or more)
non-specific reject stream (referred to streams.
as the stealth reject stream) gathers
rows from either condition to the reject
stream.

Figure 5-4 depicts how rows that are rejected by the Lookup stage are processed
by a corrective Transformer stage, where the failed references are set to a
specific value and then merged with the output of the Lookup stage.
Parallel
lnk_lkup_reference
lnk_lkup_output lnk_merged_output
lnk_lkup_input
lnk_Merge_Tagged_Rejects
lnk_Validate_Something
lnk_lkup_reject
lnk_xfmr_output
xfrm_Tag_Lkup_Rejects
Figure 5-4 Error processing Lookup example

5.7 Component usage
DataStage offers a wealth of component types for building parallel ETL flows.
This section provides guidelines appropriate use of various stages when building
a parallel job flows.
5.7.1 Server Edition components

Avoid the use of Server Edition components in parallel job flows. DataStage
Server components limit overall performance of large-volume job flows because
many components such as the BASIC Transformer use interpreted pseudo-code.
In clustered and MPP environments Server Edition components only run on the
primary (conductor) node, severely impacting scalability and network resources.
The ability to use a DataStage Server component in a parallel job is intended

only as a migration option for existing DataStage Server applications that might
benefit by making use of parallel capabilities on SMP platforms.
The following DataStage Server components must be avoided in parallel job

flows:
򐂰 BASIC Transformers
򐂰 BASIC Routines
򐂰 Server shared containers
BASIC routines are still appropriate, and necessary, for the job control
components of a DataStage Job Sequence and Before/After Job Subroutines for
parallel jobs.
5.7.2 Copy stage

For complex data flows, it is best to develop a job iteratively using the Copy
stage as a placeholder. Because the Copy stage does not require an output link,
it can be used at the end of a data flow in the following circumstances:
򐂰 For simple jobs with only two stages, the Copy stage must be used as a
placeholder so that new stages can be inserted easily should future
requirements change.
򐂰 Unless the force property is set to True, a Copy stage with a single input link
and a single output link is optimized out of the final job flow at runtime.

5.7.3 Parallel datasets
When writing intermediate results between DataStage parallel jobs, always write
to parallel datasets. Datasets achieve end-to-end parallelism across job
boundaries by writing data in partitioned form, in sort order, and in native format.
Used in this manner, parallel datasets effectively establish restart points in the
event that a job (or sequence) needs to be re-run. Datasets offer parallel I/O on
read and write operations, without overhead for format or data type conversions.
There might be instances when setting the environment variable

APT_OLD_BOUNDED_LENGTH might be beneficial. When this environment
variable is set (present in the environment), varchar columns are only stored
using the actual data length. This might improve I/O performance (and reduce
disk use) when processing a large number of varchar columns with a large
maximum length and highly-variable data lengths.
Note: Because parallel datasets are platform and configuration-specific, they

must not be used for long-term archive of source data.
5.7.4 Parallel Transformer stages

The DataStage parallel Transformer stage generates C code, which is compiled
into a parallel component. For this reason, it is important to minimize the number
of Transformers, and to use other stages (such as Copy) when derivations are
not needed.
򐂰 The Copy stage must be used instead of a Transformer for simple operations:
– Job Design placeholder between stages (unless the force option is set to
True, the parallel framework optimizes this out at runtime)
– Renaming columns
– Dropping columns
򐂰 Default type conversions (see “Default and explicit type conversions” on
page 426.)
Rename, drop (if runtime column propagation is disabled), and default type
conversion can also be performed by the output mapping tab of any stage.
򐂰 Never use the BASIC Transformer stage in large-volume job flows. Instead,
user-defined functions and routines can expand Parallel Transformer
capabilities. The BASIC Transformer is intended as a temporary-use
migration choice for existing DataStage Server jobs containing complex
routines. Even then its use must be restricted and the routines must be
converted as soon as possible.

򐂰 Consider, if possible, implementing complex derivation expressions using
regular patterns by Lookup tables instead of using a Transformer with nested
derivations. For example, the derivation expression:
If A=0,1,2,3 Then B=”X” If A=4,5,6,7 Then B=”C”
In this situation, the expression could also be implemented with a lookup table
containing values in column A and the corresponding values in column B.
򐂰 Optimize the overall job flow design to combine derivations from multiple
Transformers into a single Transformer stage when possible.
򐂰 Because the Parallel Transformer is compiled, it is faster than the Interpreted
Filter and Switch stages. The only time that Filter or Switch must be used is
when the selection clauses need to be parameterized at runtime.
򐂰 The Modify stage can be used for non-default type conversions, null handling,
and character string trimming. See 9.2, “Modify stage” on page 146.
5.7.5 BuildOp stages

BuildOps can only be used in the following circumstances:
򐂰 Complex reusable logic cannot be implemented using the Transformer.
򐂰 Existing Transformers do not meet performance requirements.
As always, performance must be tested in isolation to identify specific causes of

bottlenecks.
5.8 Job design considerations for usage and impact

analysis
It is important to capture the table definitions prior to construction of a job flow so
that linkages are set up between the table definitions and the DataStage job
whenever a table definition is loaded into a stage from the repository or saved
into the repository from a stage. These relationships are the key to the
DataStage usage analysis mechanism as well as the impact analysis. In other
words, to identify which DataStage jobs in which a particular table definition is
used, that table definition is required to be loaded into the job/stage from the
repository or saved from the job/stage into the repository.

5.8.1 Maintaining JobDesign:Table definition connection
The following development procedures ensure that the linkage between job
design and table definition is properly established and maintained.
To preserve the job design:table definition connection, DataStage developers

should populate metadata on passive stages or links in the following ways:
򐂰 Performing a load to populate the stage with an existing table definition
򐂰 Performing a save from in the stage when changing or creating a new table
definition
򐂰 Performing a drag and drop from the repository window onto the link
To preserve the job design:table definition connection, DataStage developers

should not:
򐂰 In the Transformer stage, populate output links by dragging input columns to
output columns
򐂰 Delete a table definition that is attached to a job design
򐂰 Rename table definition folders (careful, names are case-sensitive)
containing table definitions that are in use. If the folder must be renamed, then
the affected definitions must be dropped and reacquired
򐂰 Manually create a table definition in the designer without a save
򐂰 Modify table definitions without a save operation
򐂰 Change the column order in a sequential table definition
Changing the name of a properly acquired table or file does not break the
metadata connection, neither does deleting it and recreating it.

5.8.2 Verifying the job design:table definition connection
To view the job design:table definition relationships in DataStage Designer, the
“Enable editing of internal references in jobs” option must be selected in the
project properties administrator client, as shown in Figure 5-5.
Figure 5-5 Enabling internal references in administrator
Once enabled, the relationships can be viewed in the stage editor on the Edit
Column panel, as shown in Figure 5-6.
Figure 5-6 Internal references in designer

6
Chapter 6. Partitioning and collecting

Partitioning parallelism is key for the scalability of DataStage (DS) parallel jobs.
Partitioners distribute rows of a single link into smaller segments that can be
processed independently in parallel. Partitioners exist before any stage that is
running in parallel. If the prior stage was running sequentially, a fan-out icon is
drawn on the link in the Designer canvas, as shown in Figure 6-1.
Stage Running Stage Running

Sequentially in Parallel
Figure 6-1 Fan-out icon

Collectors combine parallel partitions of a single link for sequential processing.
Collectors only exist before stages running sequentially and when the previous
stage is running in parallel, and are indicated by a fan-in icon as shown in
Figure 6-2.
Stage Running Stage Running

in Parallel Sequentially
Figure 6-2 Collector icon
This section provides an overview of partitioning and collecting methods, and

provides guidelines for appropriate use in job designs. It also provides tips for
monitoring jobs running in parallel.

6.1 Partition types
Though partitioning allows data to be distributed across multiple processes
running in parallel, it is important that this distribution does not violate business
requirements for accurate data processing. For this reason, separate types of
partitioning are provided for the parallel job developer.
Partitioning methods are separated into two distinct classes:

򐂰 Keyless partitioning
Keyless partitioning distributes rows without regard to the actual data values.
Separate types of keyless partitioning methods define the method of data
distribution.
򐂰 Keyed partitioning
Keyed partitioning examines the data values in one or more key columns,
ensuring that records with the same values in those key columns are
assigned to the same partition. Keyed partitioning is used when business
rules (for example, remove duplicates) or stage requirements (for example,
join) require processing on groups of related records.
The default partitioning method used when links are created is Auto partitioning.
The partitioning method is specified in the input stage properties using the
partitioning option, as shown in Figure 6-3 on page 94.
Chapter 6. Partitioning and collecting 93

Figure 6-3 Specifying partition method
6.1.1 Auto partitioning

This is the default partitioning method for newly-drawn links, Auto partitioning
specifies that the parallel framework attempts to select the appropriate
partitioning method at runtime. Based on the configuration file, datasets, and job
design (stage requirements and properties), Auto partitioning selects between
keyless (same, round-robin, entire) and keyed (hash) partitioning methods to
produce functionally correct results and, in certain cases, to improve
performance.
In the Designer canvas, links with Auto partitioning are drawn with the link icon,
Figure 6-4 Auto partitioning icon
Auto partitioning is designed to allow beginner DataStage developers to

construct simple data flows without having to understand the details of parallel
design principles. However, the Auto partitioning method might not be the most

efficient from an overall job perspective and in certain cases can lead to wrong
results.
Furthermore, the parallel framework’s ability to determine the appropriate

partitioning method depends on the information available to it. In general, Auto
partitioning ensures correct results when using built-in stages. However,
because the parallel framework has no visibility into user-specified logic (such as
Transformer or BuildOp stages) it might be necessary to specify a partitioning
method for certain stages. For example, if the logic defined in a Transformer
stage is based on a group of related records, a keyed partitioning method must
be specified to achieve correct results.
The Preserve Partitioning flag is an internal hint that Auto partitioning uses to
attempt to preserve previously ordered data (for example, on the output of a
parallel sort). This flag is set automatically by certain stages (sort, for example),
although it can be explicitly set or cleared in the advanced stage properties of a
given stage, as shown in Figure 6-5.
Figure 6-5 Preserve Partitioning option
The Preserve Partitioning flag is part of the dataset structure, and its state is
stored in persistent datasets.
There are cases when the input stage requirements prevent partitioning from
being preserved. For example, when the upstream partitioning scheme is
round-robin, but the stage at hand is a Join. In this case, the Join requires the
data to be partitioned by hash on the Join key. In these instances, if the Preserve
Partitioning flag was set, a warning is placed in the Director log indicating the
parallel framework was unable to preserve partitioning for a specified stage.
6.1.2 Keyless partitioning

Keyless partitioning methods distribute rows without examining the contents of
the data. The partitioning methods are described in Table 6-1 on page 96.

Table 6-1 Partitioning methods
Keyless Partition Method Description
Same Retains existing partitioning from previous stage.
Round-robin Distributes rows evenly across partitions, in a

round-robin partition assignment.
Random Distributes rows evenly across partitions in a random

partition assignment.
Entire Each partition receives the entire dataset.
Same partitioning
Same partitioning performs no partitioning to the input dataset. Instead, it retains
the partitioning from the output of the upstream stage, as shown in Figure 6-6.
0 1 2
3 4 5
0 1 2
3 4 5
Figure 6-6 Same partitioning
Same partitioning does not move data between partitions (or, in the case of a
cluster or grid, between servers), and is appropriate when trying to preserve the
grouping of a previous operation (for example, a parallel Sort).
In the Designer canvas, links that have been specified with Same partitioning are
drawn with a horizontal line partitioning icon, as in Figure 6-7.
Figure 6-7 Same partitioning icon
It is important to understand the impact of Same partitioning in a given data flow.

Because Same does not redistribute existing partitions, the degree of parallelism
remains unchanged.

If the upstream stage is running sequentially, Same partitioning effectively
causes a downstream Parallel stage to also run sequentially.
If you read a parallel dataset with Same partitioning, the downstream stage runs
with the degree of parallelism used to create the dataset, regardless of the
current $APT_CONFIG_FILE.
Note: Minimize the use of SAME partitioning, using only when necessary.
Round-robin partitioning
Round-robin partitioning evenly distributes rows across partitions in a
round-robin assignment, similar to dealing cards. Round-robin partitioning has a
fairly low overhead. It is shown in Figure 6-8.
Values of Key Columns

...876543210
Round Robin
6 7 8
3 4 5
0 1 2
Figure 6-8 Round-robin partitioning
Because optimal parallel processing occurs when all partitions have the same
workload, round-robin partitioning is useful for redistributing data that is highly
skewed (there are an unequal number of rows in each partition).
Random partitioning
Like Round-robin, Random partitioning evenly distributes rows across partitions,
but using a random assignment. As a result, the order that rows are assigned to
a particular partition differ between job runs.
Because the random partition number must be calculated, Random partitioning

has a slightly higher overhead than Round-robin partitioning.
Though in theory Random partitioning is not subject to regular data patterns that
might exist in the source data, it is rarely used in functional data flows because,
though it shares basic principle of Round-robin partitioning, it has a slightly larger
overhead.

Entire partitioning
Entire partitioning distributes a complete copy of the entire dataset to each
partition, and is illustrated in Figure 6-9.

...876543210
Entire
. . .
. . .
. . .
3 3 3
2 2 2
1 1 1
0 0 0
Figure 6-9 Entire partitioning
Entire partitioning is useful for distributing the reference data of a Lookup task
(this might or might not involve the Lookup stage).
On clustered and grid implementations, Entire partitioning might have a

performance impact, as the complete dataset must be distributed across the
network to each node.
6.1.3 Keyed partitioning

Keyed partitioning examines the data values in one or more key columns,
ensuring that records with the same values in those key columns are assigned to
the same partition. Keyed partitioning is used when business rules (for example,
Remove Duplicates) or stage requirements (for example, Join) require
processing on groups of related records. Keyed partitioning is described in
Table 6-2 on page 99.

Table 6-2 Keyed partitioning
Keyed Description
Partitioning
Hash Assigns rows with the same values in one or more key columns to the
same partition using an internal hashing algorithm.
Modulus Assigns rows with the same values in a single integer key column to
the same partition using a simple modulus calculation.
Range Assigns rows with the same values in one or more key columns to the
same partition using a specified range map generated by pre-reading
the dataset.
DB2 For DB2 Enterprise Server Edition with DPF (DB2/UDB) only Matches
the internal partitioning of the specified source or target table.
6.1.4 Hash partitioning

Hash partitioning assigns rows with the same values in one or more key columns
to the same partition using an internal hashing algorithm. This is depicted in
Figure 6-10.

...0321023211
Hash
0 1 2
3 1 2
0 1 2
3
Figure 6-10 Hash partitioning
If the source data values are evenly distributed in these key columns, and there
are a large number of unique values, then the resulting partitions are of relatively
equal size.

As an example of hashing, consider the sample dataset in Table 6-3.
Table 6-3 Sample hash dataset
ID LName FName Address
Ford Henry 66 Edison Avenue
Ford Clara 66 Edison Avenue
Ford Edsel 7900 Jefferson
Ford Eleanor 7900 Jefferson
Dodge Horace 17840 Jefferson
Dodge John 75 Boston Boulevard
Ford Henry 4901 Evergreen
Ford Clara 4901 Evergreen
Ford Edsel 1100 Lakeshore
10 Ford Eleanor 1100 Lakeshore
Hashing on the LName key column produces the results depicted in Table 6-4
and Table 6-5.
Table 6-4 Partition 0


In this case, there are more instances of Ford than Dodge, producing partition
skew, which would impact performance. In this example the number of unique
values limit the degree of parallelism, regardless of the actual number of nodes in
the parallel configuration file.
Using the same source dataset, hash partitioning on the LName and FName key
columns yields the distribution with a 4-node configuration file depicted in
Table 6-6, Table 6-7, Table 6-8, and Table 6-9.




In this example, the key column combination of LName and FName yields
improved data distribution and a greater degree of parallelism. Only the unique
combination of key column values appear in the same partition when used for
hash partitioning. When using hash partitioning on a composite key (more than
one key column), individual key column values have no significance for partition
assignment.

Modulus partitioning
Modulus partitioning uses a simplified algorithm for assigning related records
based on a single integer key column. It performs a modulus operation on the
data value using the number of partitions as the divisor. The remainder is used to
assign the value to a given partition:
partition = MOD (key_value / number of partitions)
Like hash, the partition size of modulus partitioning is equally distributed as long
as the data values in the key column are equally distributed.
Because modulus partitioning is simpler and faster than hash, it must be used if
you have a single integer key column. Modulus partitioning cannot be used for
composite keys, or for a non-integer key column.
Range partitioning
As a keyed partitioning method, Range partitioning assigns rows with the same
values in one or more key columns to the same partition. Given a sufficient
number of unique values, Range partitioning ensures balanced workload by
assigning an approximately equal number of rows to each partition, unlike Hash
and Modulus partitioning where partition skew is dependent on the actual data
distribution. This is depicted in Figure 6-11.

...405160543
Range
Range
Map File
0 4
1 4
0 3
Figure 6-11 Range partitioning
To achieve this balanced distribution, Range partitioning must read the dataset
twice: the first to create a Range Map file, and the second to actually partition the
data in a flow using the Range Map. A Range Map file is specific to a given
parallel configuration file.

The read twice penalty of Range partitioning limits its use to specific scenarios,
typically where the incoming data values and distribution are consistent over
time. In these instances, the Range Map file can be re-used.
It is important to note that if the data distribution changes without recreating the
Range Map, partition balance is skewed, defeating the intention of Range
partitioning. Also, if new data values are processed outside of the range of a
given Range Map, these rows are assigned to either the first or the last partition,
depending on the value.
In another scenario to avoid, if the incoming dataset is sequential and ordered on

the key columns, Range partitioning results in sequential processing.
DB2 partitioning
The DB2/UDB Enterprise Stage (or EE Stage) matches the internal database
partitioning of the source or target DB2 Enterprise Server Edition with Data
Partitioning Facility database (previously called DB2/UDB EEE). Using the
DB2/UDB Enterprise stage, data is read in parallel from each DB2 node. And, by
default, when writing data to a target DB2 database using the DB2/UDB
Enterprise stage, data is partitioned to match the internal partitioning of the target
DB2 table using the DB2 partitioning method.
DB2 partitioning can only be specified for target DB2/UDB Enterprise stages. To
maintain partitioning on data read from a DB2/UDB Enterprise stage, use Same
partitioning on the input to downstream stages.
6.2 Monitoring partitions

At runtime, the DataStage parallel framework determines the degree of
parallelism for each stage using:
򐂰 The parallel configuration file (APT_CONFIG_FILE)
򐂰 The degree of parallelism of existing source and target datasets (and, in
certain cases, databases)
򐂰 If specified, a stage’s node pool (stage/Advanced properties)
This information is detailed in the parallel job score, which is output to the
Director job log when the environment variable APT_DUMP_SCORE is set to
True. Specific details on interpreting the parallel job score can be found in
Appendix E, “Understanding the parallel job score” on page 401.

Partitions are assigned numbers, starting at zero. The partition number is
appended to the stage name for messages written to the Director log, as shown
in Figure 6-12, where the Peek stage is running with four degrees of parallelism
(partition numbers zero through 3).
Figure 6-12 Partition numbers as shown in Director log
To display row counts per partition in the Director Job Monitor window, right-click
anywhere in the window, and select the Show Instances option, as shown in
Figure 6-13. This is useful in determining the distribution across parallel
partitions (skew). In this instance, the stage named Sort_3 is running across four
partitions (x 4 next to the stage name), and each stage is processing an equal
number (12,500) of rows for an optimal balanced workload.
Figure 6-13 Director job monitor row counts by partition
Setting the environment variable APT_RECORD_COUNTS outputs the row

count per link per partition to the Director log as each stage/node completes
processing, as illustrated in Figure 6-14.
Figure 6-14 Output of APT_RECORD_COUNTS in Director log

The Dataset Management tool (available in the Tools menu of Designer or
Director) can be used to identify the degree of parallelism and number of rows
per partition for an existing persistent dataset, as shown in Figure 6-15.
Figure 6-15 Dataset Management tool
In a non-graphical way, the orchadmin command line utility on the DataStage

server can also be used to examine a given parallel dataset.

6.3 Partition methodology
Given the numerous options for keyless and keyed partitioning, the following
objectives help to form a methodology for assigning partitioning:
򐂰 Choose a partitioning method that gives close to an equal number of rows in
each partition, and which minimizes overhead.
This ensures that the processing workload is evenly balanced, minimizing
overall run time.
򐂰 The partition method must match the business requirements and stage
functional requirements, assigning related records to the same partition if
required
Any stage that processes groups of related records (generally using one or
more key columns) must be partitioned using a keyed partition method.
This includes, but is not limited to the following stages:
– Aggregator
– Change Capture
– Change Apply
– Join, Merge
– Remove Duplicates
– Sort
It might also be necessary for Transformers and BuildOps that process
groups of related records.
In satisfying the requirements of this second objective, it might not be possible
to choose a partitioning method that gives close to an equal number of rows
in each partition.
򐂰 Unless partition distribution is highly skewed, minimize repartitioning,
especially in cluster or grid configurations
Repartitioning data in a cluster or grid configuration incurs the overhead of
network transport.
򐂰 Partition method must not be overly complex
The simplest method that meets these objectives generally is the most
efficient and yields the best performance.

Using these objectives as a guide, the following methodology can be applied:
1. Start with Auto partitioning (the default)
2. Specify Hash partitioning for stages that require groups of related records
a. Specify only the key columns that are necessary for correct grouping as
long as the number of unique values is sufficient
b. Use Modulus partitioning if the grouping is on a single integer key column
c. Use Range partitioning if the data is highly skewed and the key column
values and distribution do not change significantly over time (Range Map
can be reused)
3. If grouping is not required, use round-robin partitioning to redistribute data
equally across all partitions
This is especially useful if the input dataset is highly skewed or sequential
4. Use Same partitioning to optimize end-to-end partitioning and to minimize
repartitioning
Be mindful that Same partitioning retains the degree of parallelism of the
upstream stage
In a flow, examine up-stream partitioning and sort order and attempt to
preserve for down-stream processing. This might require re-examining key
column usage in stages and re-ordering stages in a flow (if business
requirements permit).
Across jobs, persistent datasets can be used to retain the partitioning and sort
order. This is particularly useful if downstream jobs are run with the same
degree of parallelism (configuration file) and require the same partition and
sort order.

6.4 Partitioning examples
In this section, we apply the partitioning methodology defined earlier to several
example job flows.
6.4.1 Partitioning example 1: Optimized partitioning

The Aggregator stage only outputs key column and aggregate result columns. To
add aggregate columns to every detail row, a Copy stage is used to send the
detail rows to an Inner Join and an Aggregator. The output of the Aggregator is
sent to the second input of the Join. The standard solution is to Hash partition
(and Sort) the inputs to the Join and Aggregator stages as shown in Figure 6-16.
Figure 6-16 Standard Partitioning assignment
However, on closer inspection, the partitioning and sorting of this scenario can be
optimized. Because the Join and Aggregator use the same partition keys and
sort order, we can move the Hash partition and Sort before the Copy stage, and
apply Same partitioning to the downstream links, as shown in Figure 6-17.
Figure 6-17 Optimized Partitioning assignment
This example is revisited in Chapter 7, “Sorting” on page 115 because there is

one final step necessary to optimize the sorting in this example.

6.4.2 Partitioning example 2: Use of Entire partitioning
In this example, a Transformer is used to extract data from a single header row of
an input file. In the Transformer, a new output column is defined on the header
and detail links using a single constant value derivation. This column is used as
the key for a subsequent Inner Join to attach the header values to every detail
row. Using a standard solution, both inputs to the Join are Hash partitioned and
sorted on this single join column (either explicitly, or through Auto partitioning).
This is depicted in Figure 6-18.
Header
Src Out
Detail
Figure 6-18 Standard partitioning assignment for a Join stage
Although Hash partitioning guarantees correct results for stages that require
groupings of related records, it is not always the most efficient solution,
depending on the business requirements. Although functionally correct, the
solution has one serious limitation. Remembering that the degree of parallel
operation is limited by the number of distinct values, the single value join column
assigns all rows to a single partition, resulting in sequential processing.
To optimize partitioning, consider that the single header row is really a form of
reference data. An optimized solution is to alter the partitioning for the input links
to the Join stage, as depicted in Figure 6-19.
򐂰 Use round-robin partitioning on the detail input to distribute rows across all
partitions evenly.
򐂰 Use Entire partitioning on the header input to copy the single header row to all
partitions.
Header
Src Out
Detail
Figure 6-19 Optimized Partitioning assignment based on business requirements
Because we are joining on a single value, there is no need to pre-sort the input to
the Join. We revisit this in the Sorting discussion.

To process a large number of detail records, the link order of the Inner Join is
significant. The Join stage operates by reading a single row from the Left input
and reading all rows from the Right input that match the key values. For this
reason, the link order in this example must be set so that the single header row is
assigned to the Right input, and the detail rows are assigned to the Left input, as
shown in Figure 6-20.
Figure 6-20 Specifying link order in Join stage
If defined in reverse of this order, the Join attempts to read all detail rows from
the right input (because they have the same key column value) into memory.
For advanced users, there is one further detail in this example. Because the Join
waits until it receives an End of Group (new key value) or End of Data (no more
rows on the input dataset) from the Right input, the detail rows in the Left input
buffer to disk to prevent a deadlock. (See 12.4, “Understanding buffering” on
page 180). Changing the output derivation on the header row to a series of
numbers instead of a constant value establishes the End of Group and prevent
buffering to disk.

6.5 Collector types
Collectors combine parallel partitions of an input dataset (single link) into a single
input stream to a stage running sequentially. Like partitioning methods, the
collector method is defined in the stage Input/Partitioning properties for any
stage running sequentially, when the previous stage is running in parallel, as
shown in Figure 6-21.
Figure 6-21 Specifying collector method
6.5.1 Auto collector

The Auto collector reads rows from partitions in the input dataset without
blocking if a row is unavailable on a particular partition. For this reason, the order
of rows in an Auto collector is undefined, and might vary between job runs on the
same dataset. Auto is the default collector method.
6.5.2 Round-robin collector

The Round-robin collector reads rows from partitions in the input dataset by
reading input partitions in round-robin order. The Round-robin collector is
generally slower than an Auto collector because it must wait for a row to appear
in a particular partition.

However, there is a specialized example where the Round-robin collector might
be appropriate. Consider Figure 6-22, where data is read sequentially and
passed to a round-robin partitioner.
Figure 6-22 Round-robin collector example
Assuming the data is not repartitioned in the job flow and that the number of rows
is not reduced (for example, through aggregation), then a Round-robin collector
can be used before the final sequential output to reconstruct a sequential output
stream in the same order as the input data stream. This is because a
Round-robin collector reads from partitions using the same partition order that a
Round-robin partitioner assigns rows to parallel partitions.
6.5.3 Ordered collector

An Ordered collector reads all rows from the first partition, then reads all rows
from the next partition until all rows in the dataset have been collected.
Ordered collectors are generally only useful if the input dataset has been Sorted
and Range partitioned on the same key columns. In this scenario, an Ordered
collector generates a sequential stream in sort order.
6.5.4 Sort Merge collector

If the input dataset is sorted in parallel, the Sort Merge collector generates a
sequential stream of rows in globally sorted order. The Sort Merge collector
requires one or more key columns to be defined, and these must be the same
columns, in the same order, as used to sort the input dataset in parallel. Row
order is undefined for non-key columns.

6.6 Collecting methodology
Given the options for collecting data into a sequential stream, the following
guidelines form a methodology for choosing the appropriate collector type:
1. When output order does not matter, use Auto partitioning (the default)
2. When the input dataset has been sorted in parallel, use Sort Merge collector
to produce a single, globally sorted stream of rows
3. When the input dataset has been sorted in parallel and Range partitioned, the
Ordered collector might be more efficient
4. Use a Round-robin collector to reconstruct rows in input order for
Round-robin partitioned input datasets, as long as the dataset has not been
repartitioned or reduced.

7
Chapter 7. Sorting
Traditionally, the process of sorting data uses one primary key column and,
optionally, one or more secondary key columns to generate a sequential ordered
result set. The order of key columns determines the sequence and groupings in
the result set. Each column is specified with an ascending or descending sort
order. This is the method the SQL databases use for an ORDER BY clause, as
illustrated in the following example, sorting on primary key LName (ascending),
secondary key FName (descending).
The input data is shown in Table 7-1.
Table 7-1 Input data


After Sorting by LName, FName, the results are as in Table 7-2.
Table 7-2 Sort results

However, in most cases there is no need to globally sort data to produce a single
sequence of rows. Instead, sorting is most often needed to establish order in
specified groups of data. This sort can be done in parallel.
For example, the Remove Duplicates stage selects either the first or last row
from each group of an input dataset sorted by one or more key columns. Other
stages (for example, Sort Aggregator, Change Capture, Change Apply, Join,
Merge) require pre-sorted groups of related records.

7.1 Partition and sort keys
For parallel sort in DataStage (DS) parallel jobs, the following qualifications
apply:
򐂰 Partitioning is used to gather related records, assigning rows with the same
key column values to the same partition.
򐂰 Sorting is used to establish group order in each partition, based on one or
more key columns.
Note: By definition, when data is re-partitioned, sort order is not maintained.

To restore row order and groupings, a sort is required after repartitioning.
In the following example, the input dataset from Table 7-1 on page 115 is
partitioned on the LName and FName columns. Given a 4-node configuration
file, you would see the results depicted in Table 7-3, Table 7-4, Table 7-5, and



Chapter 7. Sorting 117

Applying a parallel sort to this partitioned input dataset, using the primary key
column LName (ascending) and secondary key column FName (descending),
would generate the resulting datasets depicted in Table 7-7, Table 7-8, Table 7-9,
and Table 7-10.





The partition and sort keys do not have to match. For example, secondary sort
keys can be used to establish order in a group for selection with the Remove
Duplicates stage (which can specify first or last duplicate to retain). Say that an
input dataset consists of order history based on CustID and Order Date. Using
Remove Duplicates, you can select the most recent order for a given customer.
To satisfy those requirements you could perform the following steps:

1. Partition on CustID to group related records
2. Sort on OrderDate in Descending order
3. Remove Duplicates on CustID, with Duplicate To Retain=First
7.2 Complete (Total) sort

If a single, sequential ordered result is needed, in general it is best to use a two
step process:
1. Partition and parallel Sort on key columns.
2. Use a Sort Merge collector on these same key columns to generate a
sequential, ordered result set.
This is similar to the way parallel database engines perform their parallel sort
operations.
7.3 Link sort and Sort stage

DataStage provides two methods for parallel sorts:
򐂰 Standalone Sort stage
This is used when execution mode is set to Parallel.
򐂰 Sort on a link
This is used when using a keyed input partitioning method.
By default, both methods use the same internal sort package (the tsort operator).
The Link sort offers fewer options, but is easier to maintain in a DataStage job, as
there are fewer stages on the design canvas. The Standalone sort offers more
options, but as a separate stage makes job maintenance more complicated.
In general, use the Link sort unless a specific option is needed on the
stand-alone stage. Most often, the standalone Sort stage is used to specify the
Sort Key mode for partial sorts.

7.3.1 Link sort
Sorting on a link is specified on the Input/Partitioning stage options, when
specifying a keyed partitioning method. (Sorting on a link is not available with
Auto partitioning, although the DS parallel framework might insert a Sort if
required). When specifying key columns for partitioning, the Perform Sort option
is checked. In the Designer canvas, links that have sort defined have a Sort icon
in addition to the Partitioning icon, as shown in Figure 7-1.
Figure 7-1 Link Sort Icon
Additional properties can be specified by right-clicking the key column, as shown

in Figure 7-2.
Figure 7-2 Specifying Link Sort options
Key column options let the developer specify the following options:
򐂰 Key column usage: sorting, partitioning, or both
򐂰 Sort direction: Ascending or Descending
򐂰 Case sensitivity (strings)
򐂰 Sorting character set: ASCII (default) or EBCDIC (strings)
򐂰 Position of nulls in the result set (for nullable columns)

7.3.2 Sort stage
The standalone Sort stage offers more options than the sort on a link, as
Figure 7-3 Sort stage options
Specifically, the following properties are not available when sorting on a link:
򐂰 Sort Key Mode (a particularly important performance optimization)
򐂰 Create Cluster Key Change Column
򐂰 Create Key Change Column
򐂰 Output Statistics
򐂰 Sort Utility (do not change this)
򐂰 Restrict Memory Usage
Of the options only available in the standalone Sort stage, the Sort Key Mode is
most frequently used.
Note: The Sort Utility option is an artifact of previous releases. Always specify
the DataStage Sort Utility, which is significantly faster than a UNIX sort.

7.4 Stable sort
Stable sorts preserve the order of non-key columns in each sort group. This
requires additional overhead in the sort algorithm, and thus a stable sort is
generally slower than a non-stable sort for the same input dataset and sort keys.
For this reason, disable Stable sort unless needed.
By default, the Stable sort option is disabled for sorts on a link and enabled with
the standalone Sort stage.
7.5 Subsorts
In the standalone Sort stage, the key column property, Sort Key Mode, is a
particularly powerful feature and a significant performance optimizer. It is used
when resorting a sub-grouping of a previously sorted input dataset, instead of
performing a complete sort. This subsort uses significantly less disk space and
CPU resource, and can often be performed in memory (depending on the size of
the new subsort groups).
To resort based on a sub-group, all key columns must still be defined in the Sort
stage. Re-used sort keys are specified with the “Do not Sort (Previously Sorted)”
property. New sort keys are specified with the Sort Key Mode property, as shown
in Figure 7-4.
Figure 7-4 Sort Key Mode property

To perform a sub-sort, keys with the “Do not Sort (Previously Sorted)” property
must be at the top of the list, without gaps between them. The key column order
for these keys must match the key columns and order defined in the
previously-sorted input dataset.
If the input data does not match the key column definition for a sub-sort, the job
aborts.
7.6 Automatically-inserted sorts

By default, the parallel framework inserts sort operators as necessary to ensure
correct results. The parallel job score (see Appendix E, “Understanding the
parallel job score” on page 401) can be used to identify automatically-inserted
sorts, as shown in Figure 7-5.
op1[4p] {(parallel inserted tsort operator

{key={value=LastName}, key={value=FirstName}}(0))
on nodes (
Figure 7-5 Inserted Sort operator
Typically, the parallel framework inserts sorts before any stage that requires
matched key values or ordered groupings (Join, Merge, Remove Duplicates, Sort
Aggregator). Sorts are only inserted automatically when the flow developer has
not explicitly defined an input sort.
Though ensuring correct results, inserted sorts can be a significant performance

impact if they are not necessary. There are two ways to prevent the parallel
framework from inserting an un-necessary sort:
򐂰 Insert an upstream Sort stage on each link, define all sort key columns with
the “Do not Sort (Previously Sorted)” Sort Mode key property.
򐂰 Set the environment variable APT_SORT_INSERTION_CHECK_ONLY. This
verifies sort order but does not perform a sort, aborting the job if data is not in
the required sort order.
Revisiting the partitioning examples in 6.4, “Partitioning examples” on page 108,

the environment variable $APT_SORT_INSERTION_CHECK_ONLY must be set
to prevent the DS parallel framework from inserting unnecessary sorts before the
Join stage.

7.7 Sort methodology
Using the rules and behavior outlined in the previous section, the following
methodology must be applied when sorting in a parallel data flow:
1. Start with a link sort.
2. Specify only necessary key columns.
3. Do not use Stable Sort unless needed.
4. Use a stand-alone Sort stage instead of a Link sort for options that are not
available on a Link sort:
– Sort Key Mode, Create Cluster Key Change Column, Create Key Change
Column, Output Statistics
– Always specify the DataStage Sort Utility for standalone Sort stages
– Use the “Do not Sort (Previously Sorted)” Sort Key Mode to resort a
sub-group of a previously-sorted input dataset
5. Be aware of automatically-inserted sorts.
Set $APT_SORT_INSERTION_CHECK_ONLY to verify but do not establish
required sort order.
6. Minimize the use of sorts in a job flow.
7. To generate a single, sequential ordered result set use a parallel Sort and a
Sort Merge collector
7.8 Tuning sort

Sort is a particularly expensive task which requires CPU, memory, and disk
resources.
To perform a sort, rows in the input dataset are read into a memory buffer on
each partition. If the sort operation can be performed in memory (as is often the
case with a sub-sort) no disk I/O is performed.
By default, each sort uses 20 MB of memory per partition for its memory buffer.
This value can be changed for each standalone Sort stage using the Restrict
Memory Usage option (the minimum is 1 MB/partition). On a global basis, the
APT_TSORT_STRESS_BLOCKSIZE environment variable can be use to
specify the size of the memory buffer, in MB, for all sort operators (link and
standalone), overriding any per-sort specifications.

If the input dataset cannot fit into the sort memory buffer, results are temporarily
spooled to disk in the following order:
1. Scratch disks defined in the current configuration file (APT_CONFIG_FILE) in
the sort named disk pool
2. Scratch disks defined in the current configuration file default disk pool
3. The default directory specified by the environment variable TMPDIR
4. The directory /tmp (on UNIX) or C:/TMP (on Windows) if available
The file system configuration and number of scratch disks defined in parallel
configuration file can impact the I/O performance of a parallel sort. Having a
greater number of scratch disks for each node allows the sort to spread I/O
across multiple file systems.
7.8.1 Sorts and variable-length fields

Although it is usually recommended to define the maximum length of variable
length fields, there are situations when it is better to leave their lengths unbound.
The parallel framework always allocates the space equivalent to their maximum
specified lengths. If most values are much shorter than their maximum lengths,
there will be a large amount of unused space being moved around between
operators as well as to/from datasets and fixed format files. That happens, for
instance, when an address field is defined as "varchar(500)" but most addresses
are 30 characters long.
This severely impacts the performance of sort operations: the more unused
bytes a record holds, the more unnecessary data is moved to and from scratch
space.
In those situations, it is better to leave the length of those fields unspecified, as

the records will only allocate the exact space to hold the field values.
This rule must be applied judiciously, but it may result in great performance
gains.

8
Chapter 8. File Stage usage

DataStage (DS) offers various parallel stages for reading from and writing to files.
In this chapter we provide suggestions for when to use a particular stage, and
any limitations that are associated with that stage.
A summary of the various stages are provided in Table 8-1.
Table 8-1 File stages

File stage Suggested Usage Limitations
Sequential File Read and write standard Cannot write to a single file in
files in a single format. parallel, performance penalty
of conversion, does not
support hierarchical data files.
Complex Flat File Need to read source data Cannot write in parallel;
in complex (hierarchical) performance penalty of format
format, such as mainframe conversion.
sources with COBOL
copybook file definitions.
Data Set Intermediate storage Can only be read from and

between DataStage written to by DataStage
parallel jobs. parallel jobs or orchadmin
command.

File stage Suggested Usage Limitations
File Set Need to share information Slightly higher overhead than

with external applications, dataset.
can write in parallel
(generates multiple
segment files).
SAS Parallel Need to share data with an Requires Parallel SAS, can
external Parallel SAS only be read from / written to
application. (Requires SAS by DataStage or Parallel SAS.
connectivity license for
DataStage.)
Lookup File Set Rare instances where Can only be written – contents
Lookup reference data is cannot be read or verified. Can
required by multiple jobs only be used as reference link
and is not updated on a Lookup stage.
frequently.
No parallel file stage supports update of existing records. Certain stages (parallel
dataset) support Append, to add new records to an existing file. But this is not
recommended, as it imposes risks for failure recovery.
We provide further information about these File stage types in the remaining
sections of this chapter.

8.1 Dataset usage
Parallel datasets are the persistent (on-disk) representation of the in-memory
data structures of the parallel framework. As such, datasets store data in
partitioned form, using the internal format of the parallel engine. In general,
datasets provide maximum performance for reading and writing data from disk,
as no overhead is needed to translate data to the internal parallel framework
representation.
Data is stored in datasets in fixed-length format and variable length fields are
padded up to their maximum length. This allows the parallel framework to
determine field boundaries quickly without having to scan the entire record
looking for field delimiters.
This yields the best performance when most of the fields are of fixed length and
unused positions in variable length fields tend to be minimal.
However, when the overall amount of unused space in variable length fields is
significant, the dataset advantages tend to be offset by the cost of storing that
much space. For nstance, if an address field is defined as "varchar(500)" and
most addresses are 30 characters long, there will be a significant amount of
unused space across the entire dataset. When dealing with millions or billions of
records, this cost is significant.
Under those circumstances, there are two alternatives:

1. Define the environment variable APT_OLD_BOUNDED_LENGTH
When this environment variable is set varchar columns will only be stored
using the actual data length.
2. Use Filesets instead
Filesets are stored in text format and, if defined as delimited format, only the
exact amount of space required to hold he actual values is used.
Datasets can only be read from and written to using a DataStage parallel job. If
data is to be read or written by other applications (including DS Server jobs), then
a different parallel stage such as a SequentialFile should be adopted instead.
Also, archived datasets can only be restored to DataStage instances that are on
the exact same OS platform.
Chapter 8. File Stage usage 129

8.2 Sequential File stages (Import and export)
The Sequential File stage can be used to read from or write to one or more flat
files of the same format. Unlike the Complex Flat File stage, the Sequential File
stage can only read and write data that is in flattened (row/column) format.
8.2.1 Reading from a sequential file in parallel

The ability to read sequential files in parallel in DS parallel jobs depends on the
read method and the options specified. These are depicted for sequential and
parallel read in Table 8-2.
Table 8-2 Read method options

Sequential File - Options to read Sequentially:
Read Method: Specific Files, only one file specified might be a file or named pipe
Read Method: File Pattern
Sequential File - Options to read in Parallel:
Read Method: Specific Files, only one file specified, Readers Per Node option greater
than 1 useful for SMP configurations - file may be either fixed or variable-width
Read Method: Specific Files, more than one file specified, each file specified within a
single Sequential File stage must be of the same format
Read Method: File Pattern, set environment variable

$APT_IMPORT_PATTERN_USES_FILESET
Read Method: Specific Files, Read From Multiple Nodes option is set to Yes, useful
for cluster and Grid configurations - file may only be fixed-width
When reading in parallel, input row order is not maintained across readers.
8.2.2 Writing to a sequential file in parallel

It is only possible to write in parallel from a Sequential File stage when more than
one output file is specified. In these instances, the degree of parallelism of the
write corresponds to the number of file names specified.
A better option for writing to a set of sequential files in parallel is to use the
FileSet stage. This creates a single header file (in text format) and corresponding

data files, in parallel, using the format options specified in the FileSet stage. The
FileSet stage writes in parallel.
8.2.3 Separating I/O from column import

If the sequential file input cannot be read in parallel, performance can still be
improved by separating the file I/O from the column parsing operation. In a job,
we define a single large string column for the non-parallel sequential file read,
and pass this to a Column Import stage to parse the file in parallel. The
formatting and column properties of the Column Import stage match those of the
Sequential File stage. An example is depicted in Figure 8-1.
Figure 8-1 Column Import example
This method is also useful for External Source and FTP Sequential Source
stages.
8.2.4 Partitioning sequential file reads

Care must be taken to choose the appropriate partitioning method from a
sequential file read:
򐂰 Do not read from a sequential file using SAME partitioning in the downstream
stage. Unless more than one source file is specified, SAME reads the entire
file into a single partition, making the entire downstream flow run sequentially
(unless it is later repartitioned).
򐂰 When multiple files are read by a single Sequential File stage (using multiple
files, or by using a file pattern), each file’s data is read into a separate
partition. It is important to use Round-robin partitioning (or other partitioning
appropriate to downstream components) to distribute the data in the flow
evenly.
8.2.5 Sequential file (Export) buffering

By default, the Sequential File (export operator) stage buffers its writes to
optimize performance. When a job completes successfully, the buffers are
flushed to disk. The environment variable $APT_EXPORT_FLUSH_COUNT
allows the job developer to specify how frequently (in number of rows) that the
Sequential File stage flushes its internal buffer on writes.

Setting this value to a low number (such as 1) is useful for real-time applications,
but there is a small performance penalty associated with increased I/O. It is also
important to remember that this setting applies to all Sequential File stages in the
data flow.
8.2.6 Parameterized sequential file format

The Sequential File stage supports a schema file option to specify the column
definitions and file format of the source file. Using the schema file option allows
the format of the source file to be specified at runtime, instead of statically
through table definitions.
The format of the schema file, including sequential file import/export format
properties is documented in the Orchestrate Record Schema manual, which is
included with the product documentation. This document is required, because
the Import/Export properties used by the Sequential File and Column Import
stages are not documented in the DataStage Parallel Job Developers Guide.
8.2.7 Reading and writing nullable columns

When reading from or writing to sequential files or file sets, the in-band (value)
must be explicitly defined in the extended column attributes for each nullable
column, as shown in Figure 8-2 on page 133.

Figure 8-2 Extended column metadata (nullable properties)
By default, DataStage does not allow a zero-length null_field setting for

fixed-width columns. Setting the environment variable
$APT_IMPEXP_ALLOW_ZERO_LENGTH_FIXED_NULL permits this. This
must be used with care, as poorly formatted data causes incorrect results.
8.2.8 Reading from and writing to fixed-length files

Particular attention must be taken when processing fixed-length fields using the
Sequential File stage:
򐂰 If the incoming columns are variable-length data types (for example, Integer,
Decimal, Varchar), the field width column property must be set to match the
fixed-width of the input column. Double-click the column number in the grid
dialog box to set this column property.
򐂰 If a field is nullable, you must define the null field value and length in the
nullable section of the column property. Double-click the column number in

the grid dialog box or right-click the column and select Edit Column to set
these properties.
򐂰 When writing fixed-length files from variable-length fields (for example,
Integer, Decimal, and Varchar), the field width and pad string column
properties must be set to match the fixed-width of the output column.
Double-click the column number in the grid dialog box to set this column
property.
򐂰 To display each field value, use the print_field import property. Use caution
when specifying this option as it can generate an enormous amount of detail
in the job log. All import and export properties are listed in the Import/Export
Properties chapter of the Orchestrate Operators Reference, which is included
with the product documentation..
8.2.9 Reading bounded-length VARCHAR columns

Care must be taken when reading delimited, bounded-length varchar columns
(varchars with the length option set). By default, if the source file has fields with
values longer than the maximum varchar length, these extra characters are
silently truncated.
The $APT_IMPORT_REJECT_STRING_FIELD_OVERRUNS environment

variable directs parallel jobs to reject records with strings longer than their
declared maximum column length.
8.2.10 Tuning sequential file performance

On heavily-loaded file servers or certain RAID/SAN array configurations, the
$APT_IMPORT_BUFFER_SIZE and $APT_EXPORT_BUFFER_SIZE
environment variables can be used to improve I/O performance. These settings
specify the size of the read (import) and write (export) buffer size in Kbytes, with
a default of 128 (128 K). Increasing this size can improve performance.
Finally, in certain disk array configurations, setting the environment variable

$APT_CONSISTENT_BUFFERIO_SIZE to a value equal to the read/write size in
bytes can significantly improve performance of Sequential File operations.
8.3 Complex Flat File stage

The Complex Flat File (CFF) stage can be used to read or write one or more files
in the same hierarchical format. When used as a source, the stage allows you to
read data from one or more complex flat files, including MVS™ datasets with
QSAM and VSAM files. A complex flat file can contain one or more GROUPs,

REDEFINES, or OCCURS clauses. Complex Flat File source stages execute in
parallel mode when they are used to read multiple files. You can configure the
stage to execute sequentially if it is only reading one file with a single reader.
Note: The Complex Flat File stage cannot read from sources with OCCURS
DEPENDING ON clauses. (This is an error in the DataStage documentation.)
When used as a target, the stage allows you to write data to one or more
complex flat files. It does not write to MVS datasets.
8.3.1 CFF stage data type mapping

When you work with mainframe data using the CFF stage, the data types are
mapped to internal parallel data types as depicted in Table 8-3.
Table 8-3 Data types

COBOL Type Description Size Internal Type Internal Options
S9(1-4) binary, native binary 2 bytes int16

COMP/COMP-5

COMP/COMP-5

COMP/COMP-5
9(1-4) binary, native binary 2 bytes uint16

COMP/COMP-5

COMP/COMP-5

COMP/COMP-5
X(n) character n bytes string(n)
X(n) character for filler n bytes raw(n)
X(n) varchar n bytes string(max=n)
9(x)V9(y)COMP-3 decimal (x+y)/2+1 bytes decimal[x+y,y] packed
S9(x)V9(y)COMP-3 decimal (x+y)/2+1 bytes decimal[x+y,y] packed
9(x)V9(y) display_numeric x+y bytes decimal[x+y,y] zoned

or string[x+y]

S9(x)V9(y) display_numeric x+y bytes decimal[x+y,y] zoned, trailing
or string[x+y]
S9(x)V9(y) SIGN IS display_numeric x+y bytes decimal[x+y,y] zoned, trailing

TRAILING
S9(x)V9(y) SIGN IS display_numeric x+y bytes decimal[x+y,y] zoned, leading

LEADING
S9(x)V9(y) SIGN IS display_numeric x+y+1 bytes decimal[x+y,y] separate, trailing

TRAILING
SEPARATE
S9(x)V9(y) SIGN IS display_numeric x+y+1 bytes decimal[x+y,y] separate, leading

LEADING
SEPARATE
COMP-1 float 4 bytes sfloat
COMP-2 float 8 bytes dfloat
N(n) or G(n) graphic_n, graphic_g n*2 bytes ustring[n]

DISPLAY-1
N(n) or G(n) vargraphic_g/n n*2 bytes ustring[max=n]

DISPLAY-1
Group subrec
8.4 Filesets
Filesets are a type of hybrid between datasets and sequential files.
Just like a dataset, records are stored in several partition files. Also, there is a
descriptor file that lists the paths to those partition files as well as the schema for
the records contained in them. However, data is stored in text format as in
SequentialFiles, as opposed to datasets that stored column values in internal
binary format.
Filesets are the way of writing data to text files in parallel.

They are the best alternative under one of the following two circumstances:
򐂰 When the overall amount of unused space in variable length fields is
significant.
For instance, when an address field is defined as "varchar(500)" but most
records have addresses 30 characters long.
If the Fileset is defined as delimited format, varchar columns will only be
stored using the actual data length, yielding significant savings over datasets.
򐂰 Data must be archived to be later restored to a DataStage instance on a
different OS platform.
Records are stored in text format, so writing and reading them across
platforms is not an issue.

9
Chapter 9. Transformation languages

In this chapter we discuss the transformation languages available to DS
developers. We start by describing what we consider to be the most important
one, which is the Parallel Transformer stage. It is by far the most used
transformation language in parallel jobs. The other stages also play an important
part in parallel design, such as the Modify, Filter, and Switch stages.

9.1 Transformer stage
The DataStage (DS) Parallel Transformer stage generates C++ code that is
compiled into a parallel component. For this reason, it is important to minimize
the number of Transformers, and to use other stages (such as Copy) when
derivations are not needed. See 5.7.4, “Parallel Transformer stages” on page 86
for guidelines on Transformer stage usage.
9.1.1 Transformer NULL handling and reject link

When evaluating expressions for output derivations or link constraints, the
Transformer rejects (through the reject link indicated by a dashed line) any row
that has a NULL value used in the expression. When rows are rejected by a
Transformer, entries are placed in the Director job log.
Always include reject links in a parallel Transformer. This makes it easy to identify
reject conditions (by row counts). To create a Transformer reject link in Designer,
right-click an output link and choose Convert to Reject, as in Figure 9-1.
Figure 9-1 Transformer reject link
The parallel Transformer rejects NULL derivation results (including output link
constraints) because the rules for arithmetic and string handling of NULL values
are, by definition, undefined. Even if the target column in an output derivation
allows nullable results, the Transformer rejects the row instead of sending it to the
output links.

For this reason, if you intend to use a nullable column in a Transformer derivation
or output link constraint, it must be converted from its out-of-band (internal) null
representation to an in-band (specific value) null representation using stage
variables or the Modify stage.
For example, the following stage variable expression would convert a null value
to a specific empty string:
If ISNULL(link.col) Then “” Else link.col
If an incoming column is only used in an output column mapping, the Transformer

allows this row to be sent to the output link without being rejected.
9.1.2 Parallel Transformer system variables

The system variable @ROWNUM behaves differently in the Parallel Transformer
stage than in the Server Transformer. Because the Parallel Transformer runs in
parallel, @ROWNUM is assigned to incoming rows for each partition. When
generating a sequence of numbers in parallel, or performing parallel derivations,
the system variables @NUMPARTITIONS and @PARTITIONNUM must be used.
9.1.3 Transformer derivation evaluation

Output derivations are evaluated before any type conversions on the assignment.
For example, the PadString function uses the length of the source type, not the
target. Therefore, it is important to make sure the type conversion is done before
a row reaches the Transformer.
For example, TrimLeadingTrailing(string) works only if “string” is a VarChar field.

Thus, the incoming column must be of VarChar type before it is evaluated in the
Transformer.
9.1.4 Conditionally aborting jobs

The Transformer can be used to conditionally abort a job when incoming data
matches a specific rule. Create a new output link that handles rows that match
the abort rule. In the link constraints dialog box, apply the abort rule to this output
link, and set the Abort After Rows count to the number of rows allowed before the
job must be aborted (for example, 1).
Because the Transformer aborts the entire job flow immediately, it is possible that
valid rows have not yet been flushed from sequential file (export) buffers, or
committed to database tables. It is important to set the database commit
parameters or adjust the Sequential File buffer settings (see 8.2.5, “Sequential
file (Export) buffering” on page 131.
Chapter 9. Transformation languages 141

9.1.5 Using environment variable parameters
Job parameters can be used in any derivation expression in the parallel
Transformer stage. However, if the job parameter is an Environment Variable
(starting with a $ sign in the Job Properties/Parameters list), it cannot be used
directly in the parallel Transformer.
To use an environment variable job parameter in a Parallel Transformer, define a

new job parameter and assign its default value to the environment variable.
9.1.6 Transformer decimal arithmetic

When decimal data is evaluated by the Transformer stage, there are times when
internal decimal variables need to be generated to perform the evaluation. By
default, these internal decimal variables have a precision and scale of 38 and 10.
If more precision is required, the APT_DECIMAL_INTERM_PRECISION and

APT_DECIMAL_INTERM_SCALE environment variables can be set to the
desired range, up to a maximum precision of 255 and scale of 125. By default,
internal decimal results are rounded to the nearest applicable value. The
APT_DECIMAL_INTERM_ROUND_MODE environment variable can be used to
change the rounding behavior using one of the following keywords:
򐂰 ceil
Rounds towards positive infinity. Examples:
1.4  2.1.6  -1
򐂰 floor
Rounds towards positive infinity. Examples:
1.6  1.-1.4  -2
򐂰 round_inf
Rounds or truncates towards nearest representable value, breaking ties by
rounding positive values toward positive infinity and negative values toward
negative infinity. Examples:
1.4  1, 1.5  2, -1.4  -1, -1.5  -2
򐂰 trunc_zero
Discard any fractional digits to the right of the rightmost fractional digit
supported regardless of sign. For example, if
$APT_DECIMAL_INTERM_SCALE is smaller than the results of the internal
calculation, round or truncate to the scale size. Examples:
1.56  1.5, -1.56  -1.5.

9.1.7 Optimizing Transformer expressions and stage variables
To write efficient Transformer stage derivations, it is useful to understand what
items get evaluated and when. The evaluation sequence is as follows:
Evaluate each stage variable initial value.
For each input row to process:
Evaluate each stage variable derivation value, unless the derivation is empty
For each output link:
– Evaluate the link constraint; if true
• Evaluate each column derivation value
• Write the output record
• Else skip the link
Next output link
Next input row
The stage variables and the columns in a link are evaluated in the order in which
they are displayed in the Transformer editor. Similarly, the output links are also
evaluated in the order in which they are displayed.
From this sequence, it can be seen that there are certain constructs that are
inefficient to include in output column derivations, as they are evaluated once for
every output column that uses them. Such constructs are where the same part of
an expression is used in multiple column derivations
For example, if multiple columns in output links want to use the same substring of
an input column, the following test might appear in a number of output columns
derivations:
IF (DSLINK1.col[1,3] = “001”) THEN ...
In this case, the evaluation of the substring of DSLINK1.col[1,3] is evaluated for

each column that uses it.
This can be made more efficient by moving the substring calculation into a stage
variable. By doing this, the substring is evaluated once for every input row. In this
case, the stage variable definition would as follows:
DSLINK1.col1[1,3]
Each column derivation would start with:

IF (stageVar1 = “001” THEN ...

In fact, this example can be improved further by moving the string comparison
into the stage variable. The stage variable is as follows:
IF (DSLink1.col[1,3] = “001” THEN 1 ELSE 0
Each column derivation would start as follows:

IF (stageVar1) THEN
This reduces both the number of substring functions evaluated and string
comparisons made in the Transformer.
Where an expression includes calculated constant values
For example, a column definition might include a function call that returns a
constant value, for example:
Str(“ “,20)
This returns a string of 20 spaces. In this case, the function is evaluated every
time the column derivation is evaluated. It is more efficient to calculate the
constant value just once for the whole Transformer.
This can be achieved using stage variables. This function can be moved into a
stage variable derivation. In this case, the function is still evaluated once for
every input row. The solution here is to move the function evaluation into the
initial value of a stage variable.
A stage variable can be assigned an initial value from the stage Properties dialog
box/Variables tab in the Transformer stage editor. In this case, the variable would
have its initial value set as follows:
Str(“ “,20)
Leave the derivation of the stage variable on the main Transformer page empty.
Any expression that previously used this function is changed to use the stage
variable instead.
The initial value of the stage variable is evaluated once, before any input rows are
processed. Because the derivation expression of the stage variable is empty, it is
not re-evaluated for each input row. Therefore, its value for the whole Transformer
processing is unchanged from the initial value.
In addition to a function value returning a constant value, another example is part

of an expression, as follows:
"abc" : "def"

As with the function-call example, this concatenation is evaluated every time the
column derivation is evaluated. Because the subpart of the expression is
constant, this constant part of the expression can be moved into a stage variable,
using the initial value setting to perform the concatenation just once.
Where an expression requiring a type conversion is used as a constant, or it is

used in multiple places.
For example, an expression might include something such as this:
DSLink1.col1+"1"
In this case, the 1 is a string constant. To add it to DSLink1.col1, it must be

converted from a string to an integer each time the expression is evaluated. The
solution in this case is to change the constant from a string to an integer:
DSLink1.col1+1
In this example, if DSLINK1.col1 were a string field, a conversion is required

every time the expression is evaluated. If this only appeared once in one output
column expression, this is fine. However, if an input column is used in more than
one expression, where it requires the same type conversion in each expression,
it is more efficient to use a stage variable to perform the conversion once. In this
case, you would create, for example, an integer stage variable, specify its
derivation to be DSLINK1.col1, and use the stage variable in place of
DSLink1.col1, where that conversion would have been required.
When using stage variables to evaluate parts of expressions, the data type of the
stage variable must be set correctly for that context. Otherwise, needless
conversions are required wherever that variable is used.

9.2 Modify stage
The Modify stage is the most efficient stage available, because it uses low-level
functionality that is part of every DataStage parallel component.
As noted in the previous section, the Output Mapping properties for any parallel
stage generate an underlying modify for default data type conversions, dropping
and renaming columns.
The standalone Modify stage can be used for non-default type conversions
(nearly all date and time conversions are non-default), null conversion, and string
trim. The Modify stage uses the syntax of the underlying modify operator,
documented in the Parallel Job Developers Guide, LC18-9892, as well as the
Orchestrate Operators Reference.
9.2.1 Modify and null handling

The Modify stage can be used to convert an out-of-band null value to an in-band
null representation and vice-versa.
Note: The DataStage Parallel Job Developers Guide gives incorrect syntax for
converting an out-of-band null to an in-band null (value) representation.
To convert from an out-of-band null to an in-band null (value) representation in

Modify, the syntax is as follows:
destField[:dataType] = make_null(sourceField,value)
The elements of this syntax are defined as follows:

򐂰 destField: Destination field’s name.
򐂰 dataType: The optional data type; use it if you are also converting types.
򐂰 sourceField: Source field’s name.
򐂰 value: The value of the source field when it is null.
To convert from an in-band null to an out-of-band null, the syntax is as follows:

destField[:dataType] = handle_null (sourceField,value)
The elements of this syntax are defined as follows:

򐂰 destField: Destination field’s name.
򐂰 dataType: The optional data type; use it if you are also converting types.
򐂰 sourceField: Source field’s name
򐂰 value: The value you want to represent a null in the output.

The destField is converted from an Orchestrate out-of-band null to a value of the
field’s data type. For a numeric field, the value can be a numeric value. For
decimal, string, time, date, and time stamp fields, the value can be a string.
9.2.2 Modify and string trim

The function string_trim has been added to Modify, with the following syntax:
stringField=string_trim[character, direction, justify] (string)
Use this function to remove the characters used to pad variable-length strings
when they are converted to fixed-length strings of greater length. By default,
these characters are retained when the fixed-length string is converted back to a
variable-length string.
The character argument is the character to remove. By default, this is NULL. The
value of the direction and justify arguments can be either begin or end. Direction
defaults to end, and justify defaults to begin. Justify has no affect when the target
string has variable length.
The following example removes all leading ASCII NULL characters from the
beginning of name and places the remaining characters in an output
variable-length string with the same name:
name:string = string_trim[NULL, begin](name)
The following example removes all trailing Z characters from color, and
left-justifies the resulting hue fixed-length string:
hue:string[10] = string_trim[‘Z’, end, begin](color)
9.3 Filter and Switch stages

The Filter and Switch stages evaluate their expressions at runtime for every input
row. Because it is compiled, a parallel Transformer with output link constraints is
faster than a Filter or Switch.
Use of Filter and Switch stages must be limited to instances where the entire
filter or switch expression must be parameterized at runtime. In a Parallel
Transformer, link constraint expressions, but not data, is fixed by the developer.

10
Chapter 10. Combining data

In this chapter we discuss stages related to the combination of records. Records
can be combined both horizontally and vertically. For horizontal combination (that
is, fields values from multiple records from various input links, related through a
common key, are combined into a single record image) there are three standard
stage types available, Lookup, Join, and Merge. The vertical combination, on the
other hand, is represented by the Aggregator stage. Here, multiple records from
the same input link are combined and aggregated values are appended to the
output record definition.

10.1 Lookup versus Join versus Merge
The Lookup stage is most appropriate when the reference data for all Lookup
stage s in a job is small enough to fit into available physical memory. Each lookup
reference requires a contiguous block of shared memory. If the datasets are
larger than available memory resources, the JOIN or MERGE stage must be
used.
Limit the use of database Sparse Lookups (available in the DB2 Enterprise,
Oracle Enterprise, and ODBC Enterprise stages) to scenarios where the number
of input rows is significantly smaller (for example, 1:100 or more) than the
number of reference rows. (see 13.1.7, “Database sparse lookup versus join” on
page 197).
Sparse Lookups might also be appropriate for exception-based processing when

the number of exceptions is a small fraction of the main input data. It is best to
test both the Sparse and Normal to see which actually performs best, and to
retest if the relative volumes of data change dramatically.
10.2 Capturing unmatched records from a Join

The Join stage does not provide reject handling for unmatched records (such as
in an InnerJoin scenario). If unmatched rows must be captured or logged, an
OUTER join operation must be performed. In an OUTER join scenario, all rows
on an outer link (for example, Left Outer, Right Outer, or both links in the case of
Full Outer) are output regardless of match on key values.
During an Outer Join, when a match does not occur, the Join stage inserts values
into the unmatched non-key columns using the following rules:
򐂰 If the non-key column is defined as nullable (on the Join input links), the DS
parallel framework inserts NULL values in the unmatched columns
򐂰 If the non-key column is defined as not-nullable, the parallel framework inserts
default values based on the data type. For example, the default value for an
Integer is zero, the default value for a Varchar is an empty string (“”), and the
default value for a Char is a string of padchar characters equal to the length of
the Char column.
For this reason, care must be taken to change the column properties to allow
NULL values before the Join. This is done by inserting a Copy stage and
mapping a column from NON-NULLABLE to NULLABLE.
A Transformer stage can be used to test for NULL values in unmatched columns.

In most cases, it is best to use a Column Generator to add an indicator column,
with a constant value, to each of the inner links and test that column for the
constant after you have performed the join. This isolates your match/no-match
logic from any changes in the metadata. This is also handy with Lookups that
have multiple reference links.
10.3 The Aggregator stage

In this section we discuss and describe the Aggregator stage.
10.3.1 Aggregation method

By default, the default Aggregation method is set to Hash, which maintains the
results of each key-column value/aggregation pair in memory. Because each key
value/aggregation requires approximately 2 K of memory, the Hash Aggregator
can only be used when the number of distinct key values is small and finite.
The Sort Aggregation method must be used when the number of key values is
unknown or large. Unlike the Hash Aggregator, the Sort Aggregator requires
presorted data, but only maintains the calculations for the current group in
memory.
10.3.2 Aggregation data type

By default, the output data type of a parallel Aggregator stage calculation or
recalculation column is floating point (Double). To aggregate in decimal precision,
set the “Aggregations/Default to Decimal Output” optional property in the
Aggregator stage.
You can also specify that the result of an individual calculation or recalculation is
decimal by using the optional Decimal Output sub-property.
Performance is typically better if you let calculations occur in floating point

(Double) data type and convert the results to decimal downstream in the flow. An
exception to this is financial calculations, which must be done in decimal to
preserve appropriate precision.
Chapter 10. Combining data 151

10.3.3 Performing total aggregations
The Aggregator counts and calculates based on distinct key value groupings. To
perform a total aggregation, use the stages shown in Figure 10-1:
1. Generate a single constant-value key column using the Column Generator or
an upstream Transformer.
2. Aggregate in parallel on the generated column (partition Round-robin,
aggregate on generated key column) there is no need to sort or hash-partition
the input data with only one key column value.
3. Aggregate sequentially on the generated column.
Figure 10-1 Aggregate stage example
In Figure 10-1 two Aggregators are used to prevent the sequential aggregation
from disrupting upstream processing. Therefore, they came into existence long
before DataStage (DS) itself. Those stages perform field-by-field comparisons on
two pre-sorted input datasets.
In 10.5, “Checksum” on page 155, we discuss the use of the Checksum stage,
which implements the MD5 Checksum algorithm. Although originally developed
for quite different purposes, its common use in the DataStage world tends to be
as a more efficient record comparison method.
The Slowly Changing Dimension (SCD) stage is included in this chapter as well.
It packs a considerable amount of features specifically tailored to the SCD
problem. This stage helps identify, for instance, if new records must be created or
updated.

10.4 Comparison stages
There are a number of stages that can be used to identify changes in records, as
part of incremental data warehouse loads:
򐂰 Change Capture
򐂰 Difference
򐂰 Compare
These stages are described in detail in the DataStage Parallel Job Developer
Guide, LC18-9892.
The input datasets are expected to be pre-sorted by a common key.
Comparisons are done on a column-by-column basis. The net effect in terms of

performance is a higher processing overhead as the number of attributes to be
compared and the size of the individual columns increase.
There are differences between the workings of each stage, so one type might be
better suited than the others depending on the task at hand. For instance, the
ChangeCapture stage is supposed to be used in conjunction with the
ChangeApply. The ChangeCapture produces a set of records containing what
needs to be applied by ChangeApply to the before records to produce the after
records.
The Difference stage is similar to ChangeCapture. It is seldom used and projects

tend to resort to ChangeCapture instead, as this one is closely associated with
the ChangeApply stage.

In Figure 10-2 we present an example of comparison of the results from these
three stages. The icon for the ComparisonStage is of ChangeCapture, but any of
these three stages can be used in its place, yielding different results (depending
on the specific business scenario).
Figure 10-2 Column-by-column comparison stages
This example reads data from sequential files. The data can be stored in a
database, a persistent dataset or a fileset. Note that we are not discussing the
various techniques and optimizations for the extraction of data for the before
dataset. This is left for subsequent chapters of this book.
The before dataset is typically extracted from a data mart or a data warehouse.
The after represents the set of input records for a given processing cycle.
One could implement similar logic in various ways using other standard
components, such as a normal lookup followed by a Transformer. In this
scenario, the datasets do not have to be pre-sorted. The comparison of individual
columns is implemented by expressions inside the Transformer.

An important disadvantage of the normal lookups is that they are limited to the
amount of memory available to hold the reference dataset. Stages that require
pre-sorted datasets provide for a much better scalability, as they are not
constrained by the amount of available memory.
The point here is there are several ways of implementing the same thing in
DataStage. However, there are components that are suited for specific tasks. In
the case of comparing records, the Compare, Change Capture, and Difference
stages are specific to the problem of identifying differences between records.
They reduce the coding effort and clutter.
10.5 Checksum
The Checksum stage is a new component in Information Server8.1 that
generates an MD5 Checksum value for a set of columns in a give dataset.
The RSA MD5 algorithm was originally developed and historically used for
encryption and integrity checking. However, in the realm of DataStage
applications for business intelligence and data warehousing, the most common
use tends to be record comparison for incremental loads.
In Figure 10-3 on page 156 we present an example of the Checksum stage for
record comparison. The flow assumes the Before dataset already contains a
pre-calculated checksum, most likely stored in a target DW table. A new
checksum value is calculated for incoming records (After records). The
checksum must be calculated on the exact same set of fields for both datasets.
Records from the Before and After datasets are correlated with a Left Outer Join
(the left side being the After dataset). The subsequent Transformer contains the
logic to compare the checksums and route records to appropriate datasets.
There are a few aspects to consider:

򐂰 Checksum values must be calculated for incoming records.
򐂰 Checksum values must be pre-calculated and possibly pre-stored for the
before dataset in a source dataset or database.
򐂰 Other applications that update the same target tables must use the same
algorithm as the Checksum stage. That is, the RSA MD5 algorithm.
򐂰 The checksum values are specific to a set of attributes taken as input to the
MD5 algorithm. If the set of attributes relevant to the checksum calculation
changes, all pre-calculated checksum values must be re-calculated.

Contains a pre-
Correlates existing
calculated Checksum
and new records by
means of a common
key (Left Outer Join)
LEFT SIDE to the

Outer Join
Does the comparison on the

existing and new checksum
Creates a new values, directing records to
Checksum for the appropriate output links
incoming records
Figure 10-3 Using the Checksum stage to identify dataset differences
The Checksum-based approach tends to be more efficient than the

column-by-column comparison.
The approach to use depends on how frequently the set of relevant attributes
changes and number and size of columns involved in the comparison.
10.6 SCD stage

The SCD stage is another stage that in our opinion falls into the category of
identifying changes.
It actually contains more capabilities than to compare records. It has features

that, as the name says, are specifically tailored to the processing of slowly
changing dimensions:
򐂰 Ability to process SCDs type 1 and 2
򐂰 Supports special fields:
– Business Keys
– Surrogate Keys
– Current Indicators
– Effective and Expiration Date fields

򐂰 There are two input links (one for input data, one for reference [dimension]
records) and two output links (one for fact records and another for updates to
the dimension table).
The functionality of the SCD stage can be implemented by using other

DataStage stages, such as Lookups, Transformers, and Surrogate Key
Generators. However, the SCD stage combines features into a single
component. The net effect from a job design perspective is less clutter and ease
of development.
You can find a detailed description of the SCD stage in the InfoSphere
DataStage 8.1 Parallel Job Developer Guide, LC18-9892. The SCD stage is
Figure 10-4 The SCD stage
The input stream link contains records from which data for both Fact and
Dimension tables are derived. There is one output link for fact records and a
second output link for updates to the dimension table.
The entire set of reference records is read from the reference link and stored in
memory in the form of a lookup table. This is done at initialization. The reference
data remains in memory throughout the duration of the entire job.
This reference table cannot be refreshed dynamically. As a result, the SCD

cannot be used in real-time jobs. Because the reference data is kept entirely in
memory, it is important to restrict the set of reference records for a given job run.
This release of the Data Flow Standard Practices proposes techniques for batch
incremental loads that include a discussion on how to restrict reference datasets
(See Chapter 15, “Batch data flow design” on page 259).

That technique consists of the following elements:
򐂰 Loading a set of unique source system keys into a target temp table.
򐂰 A subsequent job extracts to a dataset the result of a database join between
the temp and the target table:
– This extracted dataset containing the result of the join serves as the
reference data for Lookups, Joins, ChangeCapture or, in this case, the
SCD stage
– The Database Join reduces significantly the set of reference records, to
only what is relevant for a given run.
There is a chance that this technique will not make the reference dataset small
enough to fit in physical memory. That depends on how much memory is
available and how large are the data volumes involved.
Avoid overflowing the physical memory when running parallel jobs. This must be
avoided for any type of application, not only DataStage jobs. There must be
enough room for all processes (OS, DataStage jobs and other applications) and
user data such lookup tables to fit comfortably in physical memory.
The moment the physical memory fills up, the operating system starts paging in
and out to swap space, which is bad for performance.
Even more grave is when the entire swap space is consumed. That is when
DataStage starts throwing fatal exceptions related to fork() errors.
As previously stated, the SCD packs in lots of features that make slowly
changing dimension easier, so we are not necessarily ruling out its use.
One might consider making use of the SCD stage when the following
circumstances are in affect:
򐂰 The SCD stage meets the business requirements (that is, the processing of
slowly changing dimensions).
򐂰 It is guaranteed the reference dataset fits in memory
– Use the techniques outlined in Chapter 15, “Batch data flow design” on
page 259.
• Avoid extracting and loading the entire contents of a dimension table
• Make sure only the relevant reference subset is extracted and loaded
into memory through the “reference link”.
If there is a possibility the reference data is too large to fit in memory, another
technique, sorted joins, must be adopted instead.

11
Chapter 11. Restructuring data

In this chapter, assume you are familiar with the most common record format in
DataStage (DS), consisting of a set of atomic fields of types such as integers,
char/varchar, float, decimal and so forth. That is the record format that most
stages understand and are what can be referred to, in relational terms, as the
First Normal Form (1NF). However, DataStage can handle record formats that
are more complex than 1NF.
One of the original purposes of the underlying DataStage parallel framework was
to parallelize COBOL applications. As such, it has been implemented with ways
of importing/exporting and representing those COBOL record formats.
Those capabilities are still present in the DataStage today.

11.1 Complex data types
DataStage can represent de-normalized data records, with fields consisting of:
򐂰 Subrecords
򐂰 Vectors
򐂰 Tagged fields (which roughly correspond to C union structures)
򐂰 Arbitrary levels of nesting of these types
Complex data types are described in detail in the follow documents:

򐂰 Orchestrate 7.5 Operators Reference
򐂰 IBM Information Server 8.1 DataStage Parallel Job Developer Guide,
LC18-9892 (Be aware that tagged fields are not described in this document.)
11.1.1 Vectors
In Figure 11-1 we show an example of a single record, whose Temps field
consists of a vector of float values. The notation used here for vector elements
consists of a integer indicating the element index, followed by a colon and the
actual value.
Id:int32 City:varchar[20] SampleDate:date Temps[]:float
0 Chicago 2009/06/01 0:78

1:80
2:76
3:65
Figure 11-1 A record with a vector field
A record might have multiple fields of vector types, as shown in Figure 11-2.
Id:int32 City:varchar[20] SampleDate:date SampleTimes[]:time Temps[]:float
0 Chicago 2009/06/01 0:06:00 0:78

1:12:00 1:80
2:18:00 2:76
3:24:00 3:65
Figure 11-2 A record with two vector fields
There is one additional field for SampleTimes. But, SampleTimes and Temps are
paired fields, so perhaps there can be a better representation.

11.1.2 Subrecords
Although valid as is, an alternate schema grouping sample times and
corresponding temperatures can be adopted. This can be accomplished by
means of a subrecord field type, as depicted in Figure 11-3.
Id:int32 City:varchar[20] SampleDate:date TempSamples[]:subrec(SampleTime[]:time;Temp:float)
0 Chicago 2009/06/01 SampleTime Temp

0:06:00 0:78
78
1:12:00 1:80
80
2:18:00 2:76
76
3:24:00 3:65
65
Figure 11-3 A record with a vector of subrecords
This method of presenting data represents an evolution of the previous two

vector scenarios, in which values previously paired are contained in subrecords.
The top level field for the vector is named TempSamples. Each subrecord in this
vector contains two fields of types time and float.
11.1.3 Tagged fields

The tagged field complex data type allows a given field to assume one of a
number of possible subrecord definitions.
One typically uses tagged fields when importing data from a COBOL data file
when the COBOL data definition contains a REDEFINES statement. A COBOL
REDEFINES statement specifies alternative data types for a single field.
Chapter 11. Restructuring data 161

In Figure 11-4 we show an example of a tagged field. The top-level field “t”
consists of subrecords, which can assume, in this example, three distinct
subrecord types.
Figure 11-4 A tagged record structure
In Figure 11-5, we show examples of records for the definition in Figure 11-4.
Each line represents a separate record. More than one record has the same
value for the key field.
Figure 11-5 Sample tagged records

You can create more complex record structures by nesting tagged fields inside
vectors of subrecords. Also, tagged fields might be of subrecord or vector type,
for example.
Tagged data types are not described by the Information Server documentation.
They are described in Orchestrate 7.5 Operators Reference. However, tagged
types can still be used by means of the Orchestrate record schema definition, as
depicted in the previous examples.
11.2 The Restructure library

DataStage supports a number of stages for the conversion of record structures.
They allow for normalization and de-normalization of vectors for instance, as well
as a combination of distinct fields into vectors and vice-versa.
In Table 11-1 we show a list of DataStage operators and corresponding stage

types.
Table 11-1 DataStage operators

Stage type Orchestrate operator
Column Import Field_import
Column Export Field_export
Make SubRecords Makesubrec
Split SubRecord Splitsubrec
Combine Records Aggtorec
Promote SubRecord Promotesubrec
Make Vector Makevect
Split Vector Splitvect
N/A Tagbatch
N/A Tagswitch

For a full detailed description of those stages and operators, see the follow
documents:
򐂰 Orchestrate 7.5 Operators Reference
򐂰 IBM Information Server 8.1 DataStage Parallel Job Developer Guide,
LC18-9892
These documents describe the corresponding stage types, with the exception of
tagbatch and tagswitch.
In Figure 11-6 we put all Restructure stages and Operators in perspective. You
can see the paths that can be followed to convert from one type to another.
The Column Import and Column Export stages can import into and export from
any of the record types.
Figure 11-6 The Restructure operators in perspective

11.2.1 Tagbatch and Tagswitch
The Restructure operators for tagged fields are seldom used and are not
exposed as DS stage types, although they are still available in the underlying
Orchestrate framework. One might use a generic stage in a DS flow to make use
of them.
When using OSH (Orchestrate Shell) scripting, Restructure operators can be

referenced as usual.
The tagswitch operator writes each tag case to a separate output link, along with
a copy of the top-level fields.
In Figure 11-7, we show the input and output schemas of a tagswitch operator.
The input schema contains a top level key field, along with three tag cases.
Instead of the nested subrecord fields being flattened out to the same output
record format, each tag case is redirected to a separate output link. The top level
fields are copied along to the corresponding output link.
tagswitch
Figure 11-7 Input and output schemas of a tagswitch operator

Figure 11-8 illustrates invoking a tagswitch operator from a DS job. The flow
includes a generic stage, which allows one to reference any underlying parallel
operator. The stage properties must include the operator name (tagswitch) and
the properties as expected by the operator argument syntax.
The record is imported using a complex record schema syntax.
Figure 11-8 Using the generic stage to Invoke the Tagswitch operator

In Figure 11-9, we present an alternative solution using standard DataStage
stages. The input records must be imported with a Type indicator. A Transformer
or a Switch stage directs each record case to a separate branch. Each branch
has a separate Column Import stage, which does the parsing of each case
separately.
Id:Int32;
The type indicator type:Int8;
must be imported fname:string;
as a column value kname:string
Id:Int32;
type:Int8;
taggedField:string
Id:Int32;
Id:Int32; type:Int8;
type:Int8; income:Int32
taggedField:string
Id:Int32;
type:Int8; Id:Int32;
Transformer (or taggedField:string type:Int8;
Switch) sends Birth:date;
each record to a Retire:date
different link,
depending on the
type
Figure 11-9 The Tagswitch functionality implemented with DS stages
The Tagbatch operator flattens the record definitions of tagged fields so all fields
from the tag cases are promoted to top level fields.

In Figure 11-10 we present an example from the Orchestrate 7.5 Operators
Reference. Each record of the input schema might assume one of three possible
formats. When running those records through a Tagbatch operator, the schema
is flattened so all fields are promoted to the top level. There are fields that
indicate the original type, named “*_present”.
Figure 11-10 Input and output schemas of a Tagbatch operator
The OSH invocation for the example is as follows:

osh "... tagbatch -key key ..."
By default, multiple records with the same key value are combined into the same
output record. Or they might be written out as separate records.

In Figure 11-11 we present a way of implementing the Tagbatch functionality
using DS stages. It is similar to the solution for Tagswitch, except that it includes
a Left Outer Join to assemble a single output record out of the multiple input tag
cases.
tagCaseA
tagCaseA_toJoin
Id:int32; Id:int32;
type:int8; type:int8;
taggedField:string fname:string;
Kname:string Id:int32;
type:int8;
toParseTagCaseA
tagCaseB_toJoin fname:string;
Id:int32; Kname:string;
Id:int32; Id:int32; tagCaseB type:int8; Income:int32;
type:int8; type:int8; Birth:date;
Income:int32 Retire:date
taggedField:string taggedField:string
toParseTagCaseB FlattenedRecord
Id:int32;
type:int8; Peek
DSLink2 Id:int32; Join_32
Sequential_File_0 SplitByType type:int8; Birth:date;
taggedField:string tagCaseC_toJoin Kname:string
toParseTagCaseC
tagCaseC
DSLink29
Id:int32; Id:int32;
Left
Remove_Duplicate_key_values
Figure 11-11 The Tagbatch Functionality Implemented with DS stages
You can see from the examples that the native Tagbatch and Tagswitch operators
provide a convenient way of dealing with tagged record structures. Instead of
having multiple stages, the flow can be simplified by using components that are
specific to the task at hand. This is often the case when dealing with complex
COBOL record types.
11.2.2 Importing complex record types

Complex values can be imported into the types described in the previous
sections. There are two ways this can be done:
򐂰 Sequential File stage
– The representation of complex record structures in the DataStage table
definition tends to be cluttered. Details such as nesting level, number of
occurrences, subrecord tagging are harder to define from scratch.
– This stage allows the use of a record schema file. With record schema
files, one can achieve reusable job designs, with a great deal of
independence from record types.

– Record schema files allow the specification of record structures to a
degree of complexity beyond what the Complex Flat File stage can
support. That is typically the case when dealing with tagged record
structures and complex binary feeds.
򐂰 Complex Flat File
– The Complex Flat File exposes the details of complex flat files as stage
properties.
– It does not allow the use of record schema files, thus job designs with this
stage is specific to a certain record definition.
– You can import complex input records to flattened record types. The input
records are multiplied depending on the nested fields and number of
occurrences.
– Complex Flat Files can be used in tandem to cascade the parsing of
nested fields.
The Sequential File stage can be used in conjunction with the Column Import
stage. This might be done for two purposes:
򐂰 De-couple the file import process from the parsing of input records: the record
parsing can be done in parallel, with a higher degree of parallelism than the
one used to read the input file.
򐂰 Simplify the parsing by using multiple column import and Transformer stages
in tandem.
The full import/export record syntax is described in the DataStage 7.5 Operators
Reference (Chapter 25: Import/Export Properties).
11.3 The Pivot Enterprise stage

The Pivot Enterprise stage is used to pivot data horizontally. The Pivot stage
maps a set of columns in an input row to a single column in multiple output rows.
You can generate a pivot index that assigns an index number to each row with a
set of pivoted data.
This stage was introduced in Information Server 8.1 and complements the set of
Restructure stages that were already present in previous releases, as discussed
in the previous sections.

In Figure 11-12 we provide examples of data before and after a horizontal pivot
operation. These examples were taken from the IBM Information Server 8.1
DataStage Parallel Job Developer Guide.
Figure 11-12 Pivot examples
One could mimic the functionality of the Pivot stage by using a combination of
Restructure stages depicted in Figure 11-6 on page 164. To go from the record
structure with fields “SampleTime0…SampleTime3” and “Temp0…3” the
following sequence of stages can be implemented:
1. MakeVector
2. MakeSubRecord
3. PromoteSubRecord
This must be done twice. First for SampleTimeN, and again for TempN.

The MakeVector stage has the following drawbacks:
򐂰 Naming requirements that lead to additional field renaming;
򐂰 It does not support the generation of Pivot Index in a straightforward way.
The Pivot stage overcomes those limitations and does the transformation from a
single record into multiple records in a single stage. It provides for a more natural
and user-friendly restructure mechanism for certain scenarios.

12
Chapter 12. Performance tuning job

designs
The ability to process large volumes of data in a short period of time depends on
all aspects of the flow and environment being optimized for maximum throughput
and performance. Performance tuning and optimization is an iterative process
that begins at job design and unit tests, proceeds through integration and volume
testing, and continues throughout an application’s production life cycle.
This section provides tips for designing a job for optimal performance, and for
optimizing the performance of a given data flow using various settings and
features in DataStage (DS).

12.1 Designing a job for optimal performance
Overall job design can be the most significant factor in data flow performance.
This section outlines performance-related tips that can be followed when building
a parallel data flow using DataStage.
򐂰 Use parallel datasets to land intermediate result between parallel jobs.
– Parallel datasets retain data partitioning and sort order, in the DS parallel
native internal format, facilitating end-to-end parallelism across job
boundaries.
– Datasets can only be read by other DS parallel jobs (or the orchadmin
command line utility). If you need to share information with external
applications, File Sets facilitate parallel I/O at the expense of exporting to
a specified file format.
򐂰 Lookup File Sets can be used to store reference data used in subsequent
jobs. They maintain reference data in DS parallel internal format, pre-indexed.
However, Lookup File Sets can only be used on reference links to a Lookup
stage. There are no utilities for examining data in a Lookup File Set.
򐂰 Remove unneeded columns as early as possible in the data flow. Every
unused column requires additional memory that can impact performance (it
also makes each transfer of a record from one stage to the next more
expensive).
– When reading from database sources, use a select list to read needed
columns instead of the entire table (if possible)
– Be alert when using runtime column propagation (RCP). It might be
necessary to disable RCP for a particular stage to ensure that columns are
actually removed using that stage’s Output Mapping.
򐂰 Specify a maximum length for Varchar columns.
Unbounded strings (Varchars without a maximum length) can have a
significant negative performance impact on a job flow.
There are limited scenarios when the memory overhead of handling large
Varchar columns would dictate the use of unbounded strings:
– Varchar columns of a large (for example, 32 K) maximum length that are
rarely populated.
– Varchar columns of a large maximum length with highly varying data sizes.

򐂰 Avoid type conversions, if possible.
When working with database sources and targets, use orchdbutil to ensure
that the design-time metadata matches the actual runtime metadata
(especially with Oracle databases).
Enable $OSH_PRINT_SCHEMAS to verify runtime schema matches job
design column definitions
Verify that the data type of defined Transformer stage variables matches the
expected result type
򐂰 Minimize the number of Transformers. For data type conversions, renaming
and removing columns, other stages (for example, Copy, Modify) might be
more appropriate. Note that unless dynamic (parameterized) conditions are
required, a Transformer is always faster than a Filter or Switch stage.
򐂰 Avoid using the BASIC Transformer, especially in large-volume data flows.
External user-defined functions can expand the capabilities of the parallel
Transformer.
򐂰 Use BuildOps only when existing Transformers do not meet performance
requirements, or when complex reusable logic is required.
Because BuildOps are built in C++, there is greater control over the efficiency
of code, at the expense of ease of development (and more skilled developer
requirements).
򐂰 Minimize the number of partitioners in a job. It is usually possible to choose a
smaller partition-key set, and to re-sort on a differing set of secondary/tertiary
keys.
򐂰 When possible, ensure data is as evenly distributed as possible. When
business rules dictate otherwise and the data volume is large and sufficiently
skewed, re-partition to a more balanced distribution as soon as possible to
improve performance of downstream stages.
򐂰 Know your data. Choose hash key columns that generate sufficient unique
key combinations (when satisfying business requirements).
򐂰 Use SAME partitioning carefully. Remember that SAME maintains the degree
of parallelism of the upstream operator.
Chapter 12. Performance tuning job designs 175

򐂰 Minimize and combine use of Sorts where possible.
– It is frequently possible to arrange the order of business logic in a job flow
to make use of the same sort order, partitioning, and groupings.
– If data has already been partitioned and sorted on a set of key columns,
specifying the Do not sort, previously sorted option for those key
columns in the Sort stage reduces the cost of sorting and take greater
advantage of pipeline parallelism.
– When writing to parallel datasets, sort order and partitioning are
preserved. When reading from these datasets, try to maintain this sorting,
if possible, by using SAME partitioning.
– The stable sort option is much more expensive than non-stable sorts, and
can only be used if there is a need to maintain an implied (that is, not
explicitly stated in the sort keys) row order.
– Performance of individual sorts can be improved by increasing the
memory usage per partition using the Restrict Memory Usage (MB)
option of the standalone Sort stage. The default setting is 20 MB per
partition. In addition, the APT_TSORT_STRESS_BLOCKSIZE
environment variable can be used to set (in units of MB) the size of the
RAM buffer for all sorts in a job, even those that have the Restrict
Memory Usage option set.
12.2 Understanding operator combination

At runtime, the DataStage parallel framework analyzes a given job design and
uses the parallel configuration file to build a job score, which defines the
processes and connection topology (datasets) between them used to execute
the job logic.
When composing the score, the DS parallel framework attempts to reduce the
number of processes by combining the logic from two or more stages (operators)
into a single process (per partition). Combined operators are generally adjacent
to each other in a data flow.
The purpose behind operator combination is to reduce the overhead associated

with an increased process count. If two processes are interdependent (one
processes the output of the other) and they are both CPU-bound or I/O-bound,
there is nothing to be gained from pipeline partitioning.

Exception: One exception to this guideline is when operator combination
generates too few processes to keep the processors busy. In these
configurations, disabling operator combination allows CPU activity to be
spread across multiple processors instead of being constrained to a single
processor. As with any example, test your results in your environment.
However, the assumptions used by the parallel framework optimizer to determine

which stages can be combined might not always be the most efficient. It is for this
reason that combination can be enabled or disabled on a per-stage basis, or
globally.
When deciding which operators to include in a particular combined operator (also

referred to as a Combined Operator Controller), the framework includes all
operators that meet the following rules:
򐂰 Must be contiguous
򐂰 Must be the same degree of parallelism
򐂰 Must be combinable.
The following is a partial list of non-combinable operators:
– Join
– Aggregator
– Remove Duplicates
– Merge
– BufferOp
– Funnel
– DB2 Enterprise stage
– Oracle Enterprise stage
– ODBC Enterprise stage
– BuildOps
In general, it is best to let the framework decide what to combine and what to
leave uncombined. However, when other performance tuning measures have
been applied and still greater performance is needed, tuning combination might
yield additional performance benefits.
There are two ways to affect operator combination:

򐂰 The environment variable APT_DISABLE_COMBINATION disables all
combination in the entire data flow, this is only recommended on pre-7.0
versions of DS.
򐂰 In Designer, combination can be set on a per-stage basis (on the
stage/Advanced tab)

The job score identifies what components are combined. (For information about
interpreting a job score dump, see Appendix E, “Understanding the parallel job
score” on page 401.) In addition, if the %CPU column is displayed in a job
monitor window in Director, combined stages are indicated by parenthesis
surrounding the %CPU, as shown in Figure 12-1.
Figure 12-1 CPU-bound combined process in job monitor
Choosing which operators to disallow combination for is as much art as science.

However, in general, it is good to separate I/O heavy operators (Sequential File,
Full Sorts, and so on.) from CPU-heavy operators (Transformer, Change
Capture, and so on). For example, if you have several Transformers and
database operators combined with an output Sequential File, it might be a good
idea to set the sequential file to be non-combinable. This prevents IO requests
from waiting on CPU to become available and vice-versa.
In fact, in this job design, the I/O-intensive FileSet is combined with a

CPU-intensive Transformer. Disabling combination with the Transformer enables
pipeline partitioning, and improves performance, as shown in this subsequent job
monitor (Figure 12-2) for the same job.
Figure 12-2 Throughput in job monitor after disabling combination

12.3 Minimizing runtime processes and resource
requirements
The architecture of the parallel framework is well-suited for processing massive
volumes of data in parallel across available resources. Toward that end,
DataStage executes a given job across the resources defined in the specified
configuration file.
There are times, however, when it is appropriate to minimize the resource

requirements for a given scenario:
򐂰 Batch jobs that process a small volume of data
򐂰 Real-time jobs that process data in small message units
򐂰 Environments running a large number of jobs simultaneously on the same
servers
In these instances, a single-node configuration file is often appropriate to

minimize job startup time and resource requirements without significantly
impacting overall performance.
There are many factors that can reduce the number of processes generated at
runtime:
򐂰 Use a single-node configuration file
򐂰 Remove all partitioners and collectors (especially when using a single-node
configuration file)
򐂰 Enable runtime column propagation on Copy stages with only one input and
one output
򐂰 Minimize join structures (any stage with more than one input, such as Join,
Lookup, Merge, Funnel)
򐂰 Minimize non-combinable stages (as outlined in the previous section) such as
Join, Aggregator, Remove Duplicates, Merge, Funnel, DB2 Enterprise, Oracle
Enterprise, ODBC Enterprise, BuildOps, BufferOp
򐂰 Selectively (being careful to avoid deadlocks) disable buffering. (Buffering is
discussed in more detail in section 12.4, “Understanding buffering” on
page 180.)

12.4 Understanding buffering
There are two types of buffering in the DS parallel framework:
򐂰 Inter-operator transport
򐂰 Deadlock prevention
12.4.1 Inter-operator transport buffering

Though it might appear so from the performance statistics, and documentation
might discuss record streaming, records do not stream from one stage to
another. They are actually transferred in blocks (as with magnetic tapes) called
Transport Blocks. Each pair of operators that have a producer/consumer
relationship share at least two of these blocks.
The first block is used by the upstream/producer stage to output data it is done
with. The second block is used by the downstream/consumer stage to obtain
data that is ready for the next processing step. After the upstream block is full and
the downstream block is empty, the blocks are swapped and the process begins
again.
This type of buffering (or record blocking) is rarely tuned. It usually only comes
into play when the size of a single record exceeds the default size of the transport
block. Setting APT_DEFAULT_TRANSPORT_BLOCK_SIZE to a multiple of (or
equal to) the record size resolves the problem. Remember, there are two of these
transport blocks for each partition of each link, so setting this value too high can
result in a large amount of memory consumption.
The behavior of these transport blocks is determined by the following

environment variables:
򐂰 APT_DEFAULT_TRANSPORT_BLOCK_SIZE
This variable specifies the default block size for transferring data between
players. The default value is 8192, with a valid value range for between 8192
and 1048576. If necessary, the value provided by a user for this variable is
rounded up to the operating system's nearest page size.
򐂰 APT_MIN_TRANSPORT_BLOCK_SIZE
This variable specifies the minimum allowable block size for transferring data
between players. The default block size is 8192. The block size cannot be less
than 8192, or greater than 1048576. This variable is only meaningful when
used in combination with APT_LATENCY_COEFFICIENT,
APT_AUTO_TRANSPORT_BLOCK_SIZE, and
APT_MAX_TRANSPORT_BLOCK_SIZE

򐂰 APT_MAX_TRANSPORT_BLOCK_SIZE
This variable specifies the maximum allowable block size for transferring data
between players. The default block size is 1048576. It cannot be less than
8192, or greater than 1048576. This variable is only meaningful when used in
combination with APT_LATENCY_COEFFICIENT,
APT_AUTO_TRANSPORT_BLOCK_SIZE, and
APT_MMIN_TRANSPORT_BLOCK_SIZE
򐂰 APT_AUTO_TRANSPORT_BLOCK_SIZE
If set, the framework calculates the block size for transferring data between
players according to the following algorithm:
If (recordSize * APT_LATENCY_COEFFICIENT <
APT_MIN_TRANSPORT_BLOCK_SIZE), then blockSize =
APT_MIN_TRANSPORT_BLOCK_SIZE
Else, if (recordSize * APT_LATENCY_COEFFICIENT >
APT_MAX_TRANSPORT_BLOCK_SIZE), then blockSize =
APT_MAX_TRANSPORT_BLOCK_SIZE
Else, blockSize = recordSize * APT_LATENCY_COEFFICIENT
򐂰 APT_LATENCY_COEFFICIENT
This variable specifies the number of records to be written to each transport
block.
򐂰 APT_SHARED_MEMORY_BUFFERS
This variable specifies the number of Transport Blocks between a pair of
operators. It must be at least two.
Note: The following environment variables are used only with fixed-length
records:
򐂰 APT_MIN/MAX_TRANSPORT_BLOCK_SIZE
򐂰 APT_LATENCY_COEFFICIENT
򐂰 APT_AUTO_TRANSPORT_BLOCK_SIZE

12.4.2 Deadlock prevention buffering
The other type of buffering, Deadlock Prevention, comes into play anytime there
is a Fork-Join structure in a job. Figure 12-3 is an example job fragment.
Figure 12-3 Fork-Join example
In this example, the Transformer creates a fork with two parallel Aggregators,
which go into an Inner Join. Note however, that Fork-Join is a graphical
description, it does not necessarily have to involve a Join stage.

To understand deadlock-prevention buffering, it is important to understand the
operation of a parallel pipeline. Imagine that the Transformer is waiting to write to
Aggregator1, Aggregator2 is waiting to read from the Transformer, Aggregator1 is
waiting to write to the Join, and Join is waiting to read from Aggregator2. This
operation is depicted in Figure 12-4. The arrows represent dependency direction,
instead of data flow.
Aggregator
1
Qu
eu
ed
Transformer Join
Waiting Waiting
Qu
e ue
d
Aggregator
2
Figure 12-4 Dependency direction
Without deadlock buffering, this scenario would create a circular dependency

where Transformer is waiting on Aggregator1, Aggregator1 is waiting on Join,
Join is waiting on Aggregator2, and Aggregator2 is waiting on Transformer.
Without deadlock buffering, the job would deadlock, bringing processing to a halt
(though the job does not stop running, it would eventually time out).

To guarantee that this problem never happens in parallel jobs, there is a
specialized operator called BufferOp. BufferOp is always ready to read or write
and does not allow a read/write request to be queued. It is placed on all inputs to
a join structure (again, not necessarily a Join stage) by the parallel framework
during job startup. So the job structure is altered to look like that in Figure 12-5.
Here the arrows now represent data-flow, not dependency.
Aggregator
BufferOp1
1
Transformer Join
Aggregator
BufferOp2
2
Figure 12-5 Data flow
Because BufferOp is always ready to read or write, Join cannot be stuck waiting
to read from either of its inputs, breaking the circular dependency and
guaranteeing no deadlock occurs.
BufferOps is also placed on the input partitions to any Sequential stage that is
fed by a Parallel stage, as these same types of circular dependencies can result
from partition-wise Fork-Joins.
By default, BufferOps allocates 3 MB of memory each (remember that this is per

operator, per partition). When that is full (because the upstream operator is still
writing but the downstream operator is not ready to accept that data yet) it begins
to flush data to the scratch disk resources specified in the configuration file
(detailed in the DataStage Designer Client Guide).
Tip: For wide rows, it might be necessary to increase the default buffer size
(APT_BUFFER_MAXIMUM_MEMORY) to hold more rows in memory.

The behavior of deadlock-prevention BufferOps can be tuned through these
environment variables:
򐂰 APT_BUFFER_DISK_WRITE_INCREMENT
This environment variable controls the blocksize written to disk as the
memory buffer fills. The default 1 MB. This variable cannot exceed two-thirds
of the APT_BUFFER_MAXIMUM_MEMORY value.
򐂰 APT_BUFFER_FREE_RUN
This environment variable details the maximum capacity of the buffer operator
before it starts to offer resistance to incoming flow, as a nonnegative (proper
or improper) fraction of APT_BUFFER_MAXIMUM_MEMORY. Values greater
than 1 indicate that the buffer operator free runs (up to a point) even when it
has to write data to disk.
򐂰 APT_BUFFER_MAXIMUM_MEMORY
This environment variable details the maximum memory consumed by each
buffer operator for data storage. The default is 3 MB.
򐂰 APT_BUFFERING_POLICY
This environment variable specifies the buffering policy for the entire job.
When it is not defined or is defined to be the null string, the default buffering
policy is AUTOMATIC_BUFFERING. Valid settings are as follows:
– AUTOMATIC_BUFFERING
Buffer as necessary to prevent dataflow deadlocks
– FORCE_BUFFERING
Buffer all virtual datasets
– NO_BUFFERING
Inhibit buffering on all virtual datasets
Attention: Inappropriately specifying NO_BUFFERING can cause dataflow

deadlock during job execution. This setting is only recommended for advanced
users.
Note: FORCE_BUFFERING can be used to reveal bottlenecks in a job design

during development and performance tuning, but degrades performance and
therefore must not be used in production job runs.

Additionally, the buffer mode, buffer size, buffer free run, queue bound, and write
increment can be set on a per-stage basis from the Input/ Advanced tab of the
stage properties, as shown in Figure 12-6.
Figure 12-6 Stage properties Input Advanced tab
Aside from ensuring that no deadlock occurs, BufferOps also have the effect of
smoothing out production/consumption spikes. This allows the job to run at the
highest rate possible even when a downstream stage is ready for data at various
times than when its upstream stage is ready to produce that data. When
attempting to address these mismatches in production/consumption, it is best to
tune the buffers on a per-stage basis, instead of globally through environment
variable settings.
Important: Choosing which stages to tune buffering for and which to leave
alone is as much art as a science, and must be considered among the last
resorts for performance tuning.

Stages upstream/downstream from high-latency stages (such as remote
databases, NFS mount points for data storage, and so forth) are a good place to
start. If that does not yield enough of a performance boost (remember to test
iteratively, changing only one thing at a time), setting the buffering policy to
FORCE_BUFFERING causes buffering to occur everywhere.
By using the performance statistics in conjunction with this buffering, you might
be able identify points in the data flow where a downstream stage is waiting on
an upstream stage to produce data. Each place might offer an opportunity for
buffer tuning.
As implied, when a buffer has consumed its RAM, it asks the upstream stage to
slow down. This is called pushback. Because of this, if you do not have force
buffering set and APT_BUFFER_FREE_RUN set to at least approximately 1000,
you cannot determine that any one stage is waiting on any other stage, as
another stage far downstream can be responsible for cascading pushback all the
way upstream to the place you are seeing the bottleneck.

13
Chapter 13. Database stage guidelines

In this chapter we present guidelines for existing database stages. That is,
non-Connector database stage types that were available in DataStage (DS) 7.X
and Information Server 8.0.1:
򐂰 Native Parallel (Enterprise) stages
򐂰 Plug-in (API) stages
Existing database stages are discussed in this chapter for ease of reference,
especially when dealing with earlier releases of Information Server (IS) and
DataStage.
Connector stages are discussed in Chapter 14, “Connector stage guidelines” on

page 221.
As a general guideline, new projects should give preference to Connector

stages, and take advantage of existing Enterprise equivalents in specific cases
where these have an edge in terms of performance.

13.1 Existing database development overview
In this section we provide guidelines for accessing any database from in
DataStage Parallel Jobs using existing database stages. Subsequent sections
provide database-specific tips and guidelines.
13.1.1 Existing database stage types

DataStage offers database connectivity through native parallel and plug-in stage
types. For certain databases (DB2, Informix, Oracle, and Teradata), multiple
stage types are available”
򐂰 Native Parallel Database stages
– DB2/UDB Enterprise
– Informix Enterprise
– ODBC Enterprise
– Oracle Enterprise
– Netezza Enterprise (load only)
– SQLServer Enterprise
– Teradata Enterprise
򐂰 Plug-In Database stages
Dynamic RDBMS
– DB2/UDB API
– DB2/UDB Load
– Informix CLI
– Informix Load
– Informix XPS Load
– Oracle OCI Load
– RedBrick Load
– Sybase IQ12 Load
– Sybase OC
– Teradata API
– Teradata MultiLoad (MultiLoad)
– Teradata MultiLoad (TPump)
Note: Not all database stages (for example, Teradata API) are visible in the
default DataStage Designer palette. You might need to customize the palette
to add hidden stages.

Native Parallel Database stages
In general, for maximum parallel performance, scalability, and features it is best
to use the native parallel database stages in a job design, if connectivity
requirements can be satisfied.
Because there are exceptions to this rule (especially with Teradata), specific
guidelines for when to use various stage types are provided in the
database-specific topics in this section.
Because of their integration with database technologies, native parallel stages

often have more stringent connectivity requirements than plug-in stages. For
example, the DB2/UDB Enterprise stage is only compatible with DB2 Enterprise
Server Edition with DPF on the same UNIX platform as the DataStage Engine.
Native parallel stages always pre-query the database for actual runtime
metadata (column names, types, attributes). This allows DataStage to match
return columns by name, not position in the stage Table Definitions. However,
care must be taken to assign the correct data types in the job design.
ODBC Enterprise stage

In general, native database components (such as the Oracle Enterprise stage)
are preferable to ODBC connectivity if both are supported on the database
platform, operating system, and version. Furthermore, the ODBC Enterprise
stage is not designed to interface with database-specific parallel load
technologies.
The benefit of ODBC Enterprise stage comes from the large number of included
and third-party ODBC drivers to enable connectivity to all major database
platforms. ODBC also provides an increased level of data virtualization that can
be useful when sources and targets (or deployment platforms) change.
DataStage bundles OEM versions of ODBC drivers from DataDirect. On UNIX,

the DataDirect ODBC Driver Manager is also included. Wire Protocol ODBC
Drivers generally do not require database client software to be installed on the
server platform.
Plug-In database stages

Plug-in stage types are intended to provide connectivity to database
configurations not offered by the native parallel stages. Because plug-in stage
types cannot read in parallel, and cannot span multiple servers in a clustered or
grid configuration, they can only be used when it is not possible to use a native
parallel stage.
From a design perspective, plug-in database stages match columns by order, not
name, so Table Definitions must match the order of columns in a query.
Chapter 13. Database stage guidelines 191

13.1.2 Database metadata
In this section we describe use and types of metadata.
Runtime metadata
At runtime, the DS native parallel database stages always pre-query the
database source or target to determine the actual metadata (column names,
data types, null-ability) and partitioning scheme (in certain cases) of the source
or target table.
For each native parallel database stage, the following elements are true:
򐂰 Rows of the database result set correspond to records of a parallel dataset.
򐂰 Columns of the database row correspond to columns of a record.
򐂰 The name and data type of each database column corresponds to a parallel
dataset name and data type using a predefined mapping of database data
types to parallel data types.
򐂰 Both the DataStage Parallel Framework and relational databases support null
values, and a null value in a database column is stored as an out-of-band
NULL value in the DataStage column.
The actual metadata used by a native parallel database stage is always

determined at runtime, regardless of the table definitions assigned by the
DataStage developer. This allows the database stages to match return values by
column name instead of position. However, care must be taken that the column
data types defined by the DataStage developer match the data types generated
by the database stage at runtime. Database-specific data type mapping tables
are included in the following sections.

Metadata import
When using the native parallel DB2 Enterprise, Informix Enterprise or Oracle
Enterprise stages, use orchdbutil to import metadata to avoid type conversion
issues. This utility is available as a server command line utility and in Designer by
clicking Table Definitions  Import  Orchestrate Schema Definitions, and
selecting the Import from Database Table option in the wizard, as illustrated in
Figure 13-1 Orchdbutil metadata import
One disadvantage to the graphical orchdbutil metadata import is that the user
interface requires each table to be imported individually.
When importing a large number of tables, it is easier to use the corresponding

orchdbutil command-line utility from the DataStage server machine. As a
command, orchdbutil can be scripted to automate the process of importing a
large number of tables.

Defining metadata for database functions
When using database functions in a SQL SELECT list in a Read or Lookup, it is
important to use SQL aliases to name the calculated columns so that they can be
referenced in the DataStage job. The alias names must be added to the Table
Definition in DataStage.
For example, the following SQL assigns the alias “Total” to the calculated column:
SELECT store_name, SUM(sales) Total
FROM store_info
GROUP BY store_name
In many cases it might be more appropriate to aggregate using the Parallel

Aggregator stage. However, there might be cases where user-defined functions
or logic need to be executed on the database server.
13.1.3 Optimizing select lists

For best performance and optimal memory usage, it is best to specify column
names on all source database stages, instead of using an unqualified Table or
SQL SELECT * read. For the Table read method, always specify the Select List
subproperty. For Auto-Generated SQL, the DataStage Designer automatically
populates the select list based on the stage’s output column definition.
The only exception to this rule is when building dynamic database jobs that use
runtime column propagation to process all columns in a source table.

13.1.4 Testing database connectivity
The View Data button on the Output/Properties tab of source database stages
lets you verify database connectivity and settings without having to create and
run a job. Test the connection using View Data button. If the connection is
successful, you see a window with the result columns and data, similar to
Figure 13-2.
Figure 13-2 Sample view: Data output
If the connection fails, an error message might appear. You are prompted to view
additional detail. Clicking YES, as in Figure 13-3, displays a detailed dialog box
with the specific error messages generated by the database stage that can be
useful in debugging a database connection failure.
Figure 13-3 View additional error detail

13.1.5 Designing for restart
To enable restart of high-volume jobs, it is important to separate the
transformation process from the database write (Load or Upsert) operation. After
transformation, the results must be landed to a parallel dataset. Subsequent jobs
should read this dataset and populate the target table using the appropriate
database stage and write method.
As a further optimization, a Lookup stage (or Join stage, depending on data

volume) can be used to identify existing rows before they are inserted into the
target table.
13.1.6 Database OPEN and CLOSE commands

The native parallel database stages provide options for specifying OPEN and
CLOSE commands. These options allow commands (including SQL) to be sent to
the database before (OPEN) or after (CLOSE) all rows are read/written/loaded to the
database. OPEN and CLOSE are not offered by plug-in database stages. For
example, the OPEN command can be used to create a temporary table, and the
CLOSE command can be used to select all rows from the temporary table and
insert into a final target table.
As another example, the OPEN command can be used to create a target table,
including database-specific options (such as table space, logging, and
constraints) not possible with the Create option. In general, it is not a good idea
to let DataStage generate target tables unless they are used for temporary
storage. There are limited capabilities to specify Create table options in the
stage, and doing so might violate data-management (DBA) policies.
It is important to understand the implications of specifying a user-defined OPEN

and CLOSE command. For example, when reading from DB2, a default OPEN
statement places a shared lock on the source. When specifying a user-defined
OPEN command, this lock is not sent and must be specified explicitly if
appropriate.
Further details are outlined in the respective database sections of the

Orchestrate Operators Reference, which is part of the Orchestrate OEM
documentation.

13.1.7 Database sparse lookup versus join
Data read by any database stage can serve as the reference input to a Lookup
Operation. By default, this reference data is loaded into memory like any other
reference link (Normal Lookup).
When directly connected as the reference link to a Lookup stage, the DB2/UDB
Enterprise, ODBC Enterprise, and Oracle Enterprise stages allow the Lookup
type to be changed to Sparse, sending individual SQL statements to the
reference database for each incoming Lookup row. Sparse Lookup is only
available when the database stage is directly connected to the reference link,
with no intermediate stages.
Important: The individual SQL statements required by a Sparse Lookup are

an expensive operation from a performance perspective. In most cases, it is
faster to use a DataStage JOIN stage between the input and DB2 reference
data than it is to perform a Sparse Lookup.
For scenarios where the number of input rows is significantly smaller (for
example, 1:100 or more) than the number of reference rows in a DB2 or Oracle
table, a Sparse Lookup might be appropriate.
13.1.8 Appropriate use of SQL and DataStage

When using relational database sources, there is often a functional overlap
between SQL and DataStage functionality. Although it is possible to use either
SQL or DataStage to solve a given business problem, the optimal
implementation involves making use of the strengths of each technology to
provide maximum throughput and developer productivity.
Though there are extreme scenarios when the appropriate technology choice is
clearly understood, there might be less obvious areas where the decision must
be made based on factors such as developer productivity, metadata capture and
re-use, and ongoing application maintenance costs. The following guidelines can
assist with the appropriate use of SQL and DataStage technologies in a given job
flow:
򐂰 When possible, use a SQL filter (WHERE clause) to limit the number of rows
sent to the DataStage job. This minimizes impact on network and memory
resources, and makes use of the database capabilities.
򐂰 Use a SQL Join to combine data from tables with a small number of rows in
the same database instance, especially when the join columns are indexed. A
join that reduces the result set significantly is also often appropriate to do in
the database.

򐂰 When combining data from large tables, or when the source includes a large
number of database tables, the efficiency of the DS parallel Sort and Join
stages can be significantly faster than an equivalent SQL query. In this
scenario, it can still be beneficial to use database filters (WHERE clause) if
appropriate.
򐂰 Avoid the use of database stored procedures (for example, Oracle PL/SQL)
on a per-row basis in a high-volume data flow. For maximum scalability and
parallel performance, implement business rules using native parallel
DataStage components.
13.2 Existing DB2 guidelines

In this section we provide guidelines for working with stages.
13.2.1 Existing DB2 stage types

DataStage provides access to DB2 databases using one of 5 stages,
summarized in the Table 13-1.
Table 13-1 Database access using stages

DataStage Stage DB2 Supports Parallel Parallel Parallel SQL
Stage name type requirement partitioned read? write? Sparse Open /
DB2? Lookup Close
DB2/UDB Native DPF, same Yes / directly Yes Yes Yes Yes
Enterprise Parallel platform as to each DB2
ETL server a node
DB2/UDB Plug-In Any DB2 Yes / No Possible No No
API through DB2 through DB2 Limitations
Client or node 0
DB2-Connect
DB2/UDB Plug-In Subject to No No No No No
Load DB2 Loader
Limitations
ODBC Native Any DB2 Yes / Nob No No No
Enterprise through DB2 through DB2
Client or node 0
DBE-Connect
Dynamic Plug-In Any DB2 Yes / No Possible No No
RDBMS through DB2 through DB2 Limitations
Client or node 0
DB2-Connect

a. It is possible to connect the DB2 UDB stage to a remote database by cataloging the remote
database in the local instance and then using it as though it were a local database. This only works
when the authentication mode of the database on the remote instance is set to client
authentication. If you use the stage in this way, you might experience data duplication when working
in partitioned instances because the node configuration of the local instance might not be the same
as the remote instance. For this reason, the client authentication configuration of a remote instance
is not recommended.
b. A patched version of the ODBC Enterprise stage allowing parallel read is available from IBM
InfoSphere Support for certain platforms. Check with IBM InfoSphere Support for availability.
For specific details on the stage capabilities, consult the DataStage

documentation (DataStage Parallel Job Developers Guide, and DataStage
Plug-In guides)
DB2/UDB Enterprise stage

DataStage provides native parallel read, lookup, upsert, and load capabilities to
parallel DB2 databases on UNIX using the native parallel DB2/UDB Enterprise
stage. The DB2/UDB Enterprise stage requires DB2 Enterprise Server Edition on
UNIX with the Data Partitioning Facility (DPF) option. (Before DB2 V8, this was
called DB2 EEE.) Furthermore, the DB2 hardware/UNIX/software platform must
match the hardware/software platform of the DataStage ETL server.
As a native, parallel component the DB2/UDB Enterprise stage is designed for

maximum performance and scalability. These goals are achieved through tight
integration with the DB2 Relational Database Management System (RDBMS),
including direct communication with each DB2 database node, and reading from
and writing to DB2 in parallel (where appropriate), using the same data
partitioning as the referenced DB2 tables.
ODBC and DB2 plug-In stages

The ODBC Enterprise and plug-in stages are designed for lower-volume access
to DB2 databases without the DPF option installed (prior to V8, DB2 EEE).
These stages also provide connectivity to non-UNIX DB2 databases, databases
on UNIX platforms that differ from the platform of the DataStage Engine, or DB2
databases on Windows or Mainframe platforms (except for the Load stage
against a mainframe DB2 instance, which is not supported).
By facilitating flexible connectivity to multiple types of remote DB2 database

servers, the use of DataStage plug-in stages limits overall performance and
scalability. Furthermore, when used as data sources, plug-in stages cannot read
from DB2 in parallel.
Using the DB2/UDB API stage or the Dynamic RDBMS stage, it might be
possible to write to a DB2 target in parallel, because the DataStage parallel
framework instantiates multiple copies of these stages to handle the data that

has already been partitioned in the parallel framework. Because each plug-in
invocation opens a separate connection to the same target DB2 database table,
the ability to write in parallel might be limited by the table and index configuration
set by the D2 database administrator.
The DB2/API (plug-in) stage must be used to read from and write to DB2
databases on non-UNIX platforms (such as mainframe editions through
DB2-Connect). Sparse Lookup is not supported through the DB2/API stage.
13.2.2 Connecting to DB2 with the DB2/UDB Enterprise stage

Perform the following steps to connect to DB2 with the DB2/UDB Enterprise
stage:
1. Create a Parallel job and add a DB2/UDB Enterprise stage.
2. Add the properties as depicted in Figure 13-4.
Figure 13-4 DB2/UDB Enterprise stage properties

3. For connection to a remote DB2/UDB instance, set the following properties on
the DB2/UDB Enterprise stage in your parallel job:
– Client Instance Name: Set this to the DB2 client instance name. If you set
this property, DataStage assumes you require remote connection.
– Server (Optional): Set this to the instance name of the DB2 server.
Otherwise use the DB2 environment variable, DB2INSTANCE, to identify
the instance name of the DB2 server.
– Client Alias DB Name: Set this to the DB2 client’s alias database name for
the remote DB2 server database. This is required only if the client’s alias is
different from the actual name of the remote server database.
– Database (Optional): Set this to the remote server database name.
Otherwise use the environment variables $APT_DBNAME or
$APT_DB2DBDFT to identify the database.
– User: Enter the user name for connecting to DB2, this is required for a
remote connection to retrieve the catalog information from the local
instance of DB2 and thus must have privileges for that local instance.
– Password: Enter the password for connecting to DB2, this is required for a
remote connection to retrieve the catalog information from the local
instance of DB2 and thus must have privileges for that local instance.
13.2.3 Configuring DB2 multiple instances in one DataStage job

Although it is not officially supported, it is possible to connect to more than one
DB2 instance in a single job. Your job must meet one of the following
configurations (The use of the word stream refers to a contiguous flow of one
stage to another in a single job):
򐂰 Single stream - Two Instances Only
Reading from one instance and writing to another instance with no other DB2
instances (not sure how many stages of these two instances can be added to
the canvas for this configuration for lookups)
򐂰 Two Stream – One Instance per Steam
Reading from instance A and writing to instance A and reading from instance
B and writing to instance B (not sure how many stages of these two instances
can be added to the canvas for this configuration for lookups)
򐂰 Multiple Stream - with N DB2 sources with no DB2 targets
Reading from one to n DB2 instances in separate source stages with no
downstream other DB2 stages

To get this configuration to work correctly, you must adhere to all of the directions
specified for connecting to a remote instance AND the following:
򐂰 You must not set the APT_DB2INSTANCE_HOME environment variable.
After this variable is set, it tries to use it for each of the connections in the job.
Because a db2nodes.cfg file can only contain information for one instance,
this creates problems.
򐂰 In order for DataStage to locate the db2nodes.cfg file, you must build a user
on the DataStage server with the same name as the instance to which you
are trying to connect. (The default logic for the DB2/UDB Enterprise stage is
to use the home directory of the instance, as defined for the UNIX user with
the same name as the DB2 instance.) In the user’s UNIX home directory,
create a sqllib subdirectory and place the db2nodes.cfg file of the remote
instance there. Because the APT_DB2INSTANCE_HOME is not set, DS
defaults to this directory to find the configuration file for the remote instance.
To connect to multiple DB2 instances, use separate jobs with their respective
DB2 environment variable settings, landing intermediate results to a parallel
dataset. Depending on platform configuration and I/O subsystem performance,
separate jobs can communicate through named pipes, although this incurs the
overhead of Sequential File stage (corresponding export/import operators),
which does not run in parallel.
If the data volumes are sufficiently small, DB2 plug-in stages (DB2 API, DB2
Load, Dynamic RDBMS) can be used to access data in other instances.
13.2.4 DB2/UDB Enterprise stage column names

At runtime, the native parallel DB2/UDB Enterprise stage translates column
names exactly except when a component of a DB2 column name is not
compatible with DataStage column naming conventions. The naming
conventions place no limit on the length of a column name, but have the following
restrictions:
򐂰 The name must start with a letter or underscore character
򐂰 The name can contain only alphanumeric and underscore characters
򐂰 The name is case insensitive
When there is an incompatibility, the DB2/UDB Enterprise stage converts the
DB2 column name as follows:
򐂰 If the DB2 column name does not begin with a letter or underscore, the string
APT__column# (two underscores) is added to beginning of the column name,
where “column#” is the number of the column. For example, if the third DB2
column is named 7dig, the DataStage column is named APT__37dig.
򐂰 If the DB2 column name contains a character that is not alphanumeric or an
underscore, the character is replaced by two underscore characters.

13.2.5 DB2/API stage column names
When using the DB2/API, DB2 Load, and Dynamic RDBMS plug-in stages, set
the environment variable $DS_ENABLE_RESERVED_CHAR_CONVERT if your
DB2 database uses the reserved characters # or $ in column names. This
converts these special characters into an internal representation DataStage can
understand.
Observe the following guidelines when

$DS_ENABLE_RESERVED_CHAR_CONVERT is set:
򐂰 Avoid using the strings __035__ and __036__ in your DB2 column names
(these are used as the internal representations of # and $ respectively)
򐂰 Import metadata using the Plug-in Meta Data Import tool. Avoid hand editing
(this minimizes the risk of mistakes or confusion).
򐂰 Once the table definition is loaded, the internal column names are displayed
rather than the original DB2 names both in table definitions and in the data
browser. They are also used in derivations and expressions.
򐂰 The original names are used in generated SQL statements, however. Use
them if entering SQL in the job yourself.
13.2.6 DB2/UDB Enterprise stage data type mapping

The DB2 database schema to be accessed must not have any columns with User
Defined Types (UDTs). Use the db2 describe table [table-name] command on
the DB2 client for each table to be accessed to determine if UDTs are in use.
Alternatively, examine the DDL for each schema to be accessed.

Table definitions must be imported into DataStage using orchdbutil to ensure
accurate Table definitions. The DB2/UDB Enterprise stage converts DB2 data
types to DataStage parallel data types, as shown in Table 13-2.
Table 13-2 Parallel data types

DB2 data types DataStage parallel data types
CHAR(n) string[n] or ustring[n]
CHARACTER VARYING(n,r) string[max=n] or ustring[max=n]
DATE date
DATETIME Time or time stamp with corresponding fractional

precision for time:
򐂰 If the DATETIME starts with a year component, the
result is a time stamp field.
򐂰 If the DATETIME starts with an hour, the result is a
time field.
DECIMAL[p,s] decimal[p,s] where p is the precision and s is the scale
DOUBLE-PRECISION dfloat
FLOAT dfloat
INTEGER int32
MONEY decimal
NCHAR(n,r) string[n] or ustring[n]
NVARCHAR(n,r) string[max=n] or ustring[max=n]
REAL sfloat
SERIAL int32
SMALLFLOAT sfloat
SMALLINT int16
VARCHAR(n) string[max=n] or ustring[max=n]
Important: DB2 data types that are not listed in the table cannot be used in
the DB2/UDB Enterprise stage, and generate an error at runtime.

13.2.7 DB2/UDB Enterprise stage options
The DB2/UDB Enterprise (native parallel) stage must be used for reading from,
performing lookups against, and writing to a DB2 Enterprise Server Edition
database with Database Partitioning Feature (DPF)
򐂰 As a native, parallel component, the DB2/UDB Enterprise stage is designed
for maximum performance and scalability against large partitioned DB2 UNIX
databases.
򐂰 DB2/UDB Enterprise stage is tightly integrated with the DB2 RDBMS,
communicating directly with each database node, reading from and writing to
DB2 in parallel (where appropriate), and using the same data partitioning as
the referenced DB2 tables.
When writing to a DB2 database in parallel, the DB2/UDB Enterprise stage offers
the choice of SQL (insert/update/upsert/delete) or fast DB2 loader methods. The
choice between these methods depends on required performance, database log
usage, and recoverability.
򐂰 The Write Method (and corresponding insert/update/upsert/delete)
communicates directly with the DB2 database nodes to execute instructions
in parallel. All operations are logged to the DB2 database log, and the target
tables might be accessed by other users. Time and row-based commit
intervals determine the transaction size, and the availability of new rows to
other applications.
򐂰 The DB2 Load method requires that the DataStage user running the job have
DBADM privilege on the target DB2 database. During the load operation, the
DB2 Load method places an exclusive lock on the entire DB2 table space into
which it loads the data. No other tables in that table space can be accessed
by other applications until the load completes. The DB2 load operator
performs a non-recoverable load. That is, if the load operation is terminated
before it is completed, the contents of the table are unusable and the table
space is left in a load pending state. In this scenario, the DB2 Load DataStage
job must be rerun in Truncate mode to clear the load pending state.
13.2.8 Performance notes

In certain cases, when using user-defined SQL without partitioning against large
volumes of DB2 data, the overhead of routing information through a remote DB2
coordinator might be significant. In these instances, it might be beneficial to have
the DB2 DBA configure separate DB2 coordinator nodes (no local data) on each
ETL server (in clustered Extract, Transform, and Load (ETL) configurations). In
this configuration, DB2 Enterprise stage should not include the Client Instance
Name property, forcing the DB2 Enterprise stages on each ETL server to
communicate directly with their local DB2 coordinator.

13.3 Existing Informix database guidelines
In this section we describe the Informix database guidelines.
13.3.1 Informix Enterprise stage column names

For each Informix Enterprise stage, the following elements must be in effect:
򐂰 Columns of the database row correspond to columns of a DataStage parallel
record.
򐂰 Both DS parallel jobs and Informix support null values, and a null value in a
database column is stored as an out-of-band NULL value in the DS column.
13.3.2 Informix Enterprise stage data type mapping

Table Definitions must be imported into DataStage using orchdbutil to ensure
accurate Table Definitions. The Informix Enterprise stage converts Informix data
types to parallel data types, as shown in Table 13-3 on page 207.

Table 13-3 Data type mapping
Informix Data Type DS Parallel Data Type
CHAR(n) string[n]
CHARACTER VARYING(n,r) string[max=n]
DATE date
DATETIME date, time or time stamp with corresponding fractional

precision for time:
򐂰 If the DATETIME starts with a year component
and ends with a month, the result is a date field.
򐂰 If the DATETIME starts with a year component,
the result is a time stamp field.
򐂰 If the DATETIME starts with an hour, the result is a
time field.
DECIMAL[p,s] decimal[p,s] where p is the precision and s is the scale

The maximum precision is 32. A decimal with floating
scale is converted to dfloat.
DOUBLE-PRECISION dfloat
FLOAT dfloat
INTEGER int32
MONEY decimal
NCHAR(n,r) string[n]
NVARCHAR(n,r) string[max=n]
REAL sfloat
SERIAL int32
SMALLFLOAT sfloat
SMALLINT int16
VARCHAR(n) string[max=n]
Important: Informix data types that are not listed in the table cannot be used
in the Informix Enterprise stage, and generates an error at runtime.

13.4 ODBC Enterprise guidelines
In this section we describe the ODBC Enterprise guidelines.
13.4.1 ODBC Enterprise stage column names

For each ODBC Enterprise stage the following elements must be in effect:
򐂰 Columns of the database row correspond to columns of a DS record.
򐂰 Names are translated exactly except when the external data source column
name contains a character that DataStage does not support. In that case, two
underscore characters replace the unsupported character.
򐂰 Both DS and ODBC support null values, and a null value in a database
column is stored as an out-of-band NULL value in the DS column.
13.4.2 ODBC Enterprise stage data type mapping

ODBC data sources are not supported by the orcdbutil utility. It is important to
verify the correct ODBC to DS parallel data type mapping, as shown in
Table 13-4.

ODBC Data Type DS Parallel Data Type
SQL_BIGINT int64
SQL_BINARY raw(n)
SQL_CHAR string[n]
SQL_DECIMAL decimal[p,s] where p is the precision and s

is the scale
SQL_DOUBLE decimal[p,s]
SQL_FLOAT decimal[p,s]
SQL_GUID string[36]
SQL_INTEGER int32

ODBC Data Type DS Parallel Data Type
SQL_BIT int8 [0 or 1]
SQL_REAL decimal[p,s]
SQL_SMALLINT int16
SQL_TINYINT int8
SQL_TYPE_DATE date
SQL_TYPE_TIME time[p]
SQL_TYPE_TIMESTAMP timestamp[p]
SQL_VARBINARY raw[max=n]
SQL_VARCHAR string[max=n]
SQL_WCHAR ustring[n]
SQL_WVARCHAR ustring[max=n]
The maximum size of a DataStage record is limited to 32 K. If you attempt to read

a record larger than 32 K, the parallel framework returns an error and abort your
job.
Important: ODBC data types that are not listed in the table cannot be used in
the ODBC Enterprise stage, and generates an error at runtime.
13.4.3 Reading ODBC sources in parallel

Starting with DataStage release 7.5.2, the ODBC Enterprise stage can read in
parallel. By default, the ODBC Enterprise stage reads sequentially.
To read in parallel with the ODBC Enterprise stage, specify the Partition Column
option. For optimal performance, this column must be indexed in the database.
13.4.4 Writing to ODBC targets in parallel

In general, avoid writing to ODBC targets in parallel, even though the ODBC
Enterprise stage supports parallel execution.
Depending on the target database, and the table configuration (Row- or

Page-Level lock mode if available), it might be possible to write to a target
database in parallel using the ODBC Enterprise stage.

13.5 Oracle database guidelines
In this section we describe the Oracle database guidelines.
13.5.1 Oracle Enterprise stage column names

For each Oracle Enterprise stage, the following elements must be in effect:
򐂰 The name and data type of each database column corresponds to a dataset
name and data type using a predefined mapping of database data types to
parallel data types.
򐂰 Names are translated exactly except when the Oracle source column name
contains a character that DataStage does not support. In that case, two
򐂰 Both DS and Oracle support null values, and a null value in a database
13.5.2 Oracle Enterprise stage data type mapping

Oracle table definitions must be imported into DataStage using orchdbutil to
ensure accurate table definitions. This is particularly important for Oracle
databases, which are not heavily typed. DataStage maps Oracle data types
based on the rules given in the following table:
Table 13-5 Mapping data types

Oracle Data Type DS Parallel Data Type
CHAR(n) string[n] or ustring[n]

a fixed-length string with length = n
DATE timestamp
NUMBER decimal[38,10]
NUMBER[p,s] int32 if precision(p) < 11 and scale s = 0

decimal[p,s] if precision (p) >=11 or scale > 0
RAW(n) not supported
VARCHAR(n) string[max=n] or ustring[max=n]

a variable-length string with maximum length = n

The maximum size of a DataStage record is limited to 32 K. If you attempt to read
a record larger than 32 K, DataStage returns an error and aborts your job.
Important: Oracle data types that are not listed in the table cannot be used in
the Oracle Enterprise stage, and generate an error at runtime.
13.5.3 Reading from Oracle in parallel

By default, the Oracle Enterprise stage reads sequentially from its source table
or query. Setting the partition table option to the specified table enables parallel
extracts from an Oracle source. The underlying Oracle table does not have to be
partitioned for parallel reads in DS parallel jobs.
It is important to note that certain types of queries cannot run in parallel:

򐂰 Queries containing a GROUP BY clause that are also hash partitioned on the
same field
򐂰 Queries performing a non-collocated join (an SQL JOIN between two tables
that are not stored in the same partitions with the same partitioning strategy)
13.5.4 Oracle load options

When writing to an Oracle table (using Write Method = Load), the Oracle
Enterprise stage uses the Parallel Direct Path Load method. When using this
method, the Oracle stage cannot write to a table that has indexes (including
indexes automatically generated by Primary Key constraints) on it unless you
specify the Index Mode option (maintenance, rebuild).
Setting the environment variable $APT_ORACLE_LOAD_OPTIONS to

OPTIONS (DIRECT=TRUE, PARALLEL=FALSE) allows loading of indexed
tables without index maintenance. In this instance, the Oracle load is done
sequentially.
The Upsert Write Method can be used to insert rows into a target Oracle table
without bypassing indexes or constraints. To generate the SQL required by the
Upsert method, the key columns must be identified using the check boxes in the
column grid.

13.6 Sybase Enterprise guidelines
In this section we describe the Sybase Enterprise guidelines.
13.6.1 Sybase Enterprise stage column names

For each Sybase Enterprise stage, the following elements must be in effect:
򐂰 The name and data type of each database column corresponds to a dataset
name and data type using a predefined mapping of database data types to
parallel data types.
򐂰 Names are translated exactly except when the Sybase source column name
contains a character that DataStage does not support. In that case, two
򐂰 Both DS and Sybase support null values, and a null value in a database
13.6.2 Sybase Enterprise stage data type mapping

Sybase databases are not supported by the orcdbutil utility. It is important to
verify the correct Sybase to DS parallel data type mapping, as shown in
Table 13-6.

Sybase Data Types DS Parallel Data Types
BINARY(n) raw(n)
BIT int8
CHAR(n) string[n] a fixed-length string with length n
DATE date
DATETIME timestamp
DEC[p,s] or DECIMAL[p,s] decimal[p,s] where p is the precision and

s is the scale
DOUBLE PRECISION or FLOAT dfloat

Sybase Data Types DS Parallel Data Types
INT or INTEGER int32
MONEY decimal[15,4]
NCHAR(n) ustring[n] a fixed-length string with length

n - only for ASE
NUMERIC[p,s] decimal[p,s] where p is the precision and

s is the scale
NVARCHAR(n,r) ustring[max=n] a variable-length string

with length n - only for ASE
REAL sfloat
SERIAL int32
SMALLDATETIME timestamp
SMALLFLOAT sfloat
SMALLINT int16
SMALLMONEY decimal[10,4]
TINYINT int8
TIME time
UNSIGNED INT unit32
VARBINARY(n) raw[max=n]
VARCHAR(n) string[max=n] a variable-length string with

maximum length n
Important: Sybase data types that are not listed in the table cannot be used in
the Sybase Enterprise stage, and generates an error at runtime.

13.7 Existing Teradata database guidelines
In this section we describe the Teradata database guidelines.
13.7.1 Choosing the proper Teradata stage

In DataStage parallel jobs, the following stages depicted in Table can be used for
reading from and writing to Teradata databases:
򐂰 Source Teradata stages
– Teradata API
򐂰 Target Teradata stages
– Teradata API
– Teradata MultiLoad (MultiLoad option)
– Teradata MultiLoad (TPump option)
For maximum performance of high-volume data flows, the native parallel

Teradata Enterprise stage must be used. Teradata Enterprise uses the
programming interface of the Teradata utilities FastExport (reads) and FastLoad
(writes), and is subject to all these utilities’ restrictions.
Note: Unlike the FastLoad utility, the Teradata Enterprise stage supports
Append mode, inserting rows into an existing target table. This is done
through a shadow terasync table.
Teradata has a system-wide limit to the number of concurrent database utilities.

Each use of the Teradata Enterprise stages counts toward this limit.

13.7.2 Source Teradata stages
The Source Teradata stages are listed in Table 13-7.
Table 13-7 Source Teradata stages

Teradata stage Stage type Usage guidelines Parallel Teradata
read utility limit
Teradata Native 򐂰 Reading a large number of rows in Yes applies
Enterprise Parallel parallel
򐂰 Supports OPEN and CLOSE
commands
򐂰 Subject to the limits of Teradata
FastExport
Teradata API Plug-In Reading a small number of rows No none
sequentially
13.7.3 Target Teradata stages

The Target Teradata stages are listed in Table 13-8.
Table 13-8 Target Teradata stages

Teradata stage Stage Usage guidelines Parallel Teradata
type write utility limit
Teradata Native 򐂰 Writing a large number of rows in parallel Yes applies
Enterprise Parallel 򐂰 Supports OPEN and CLOSE commands
򐂰 Limited to INSERT (new table) or
APPEND (existing table)
򐂰 Subject to the limits of Teradata FastLoad
(but also supports APPEND)
򐂰 Locks the target table in exclusive mode
Teradata Plug-In 򐂰 Insert, Update, Delete, Upsert of moderate No applies
MultiLoad data volumes
(MultiLoad utility) 򐂰 Locks the target tables in exclusive mode
Teradata Plug-In 򐂰 Insert, Update, Delete, Upsert of small No applies
MultiLoad volumes of data in a large database
(TPump utility) 򐂰 Does not lock the target tables
򐂰 Must not be run in parallel, because each
node and use counts toward system-wide
Teradata utility limit
Teradata API Plug-In 򐂰 Insert, Update, Delete, Upsert of small Yes none
volumes of data
򐂰 Allows concurrent writes (does not lock
target)
򐂰 Slower than TPump for equivalent
operations

13.7.4 Teradata Enterprise stage column names
For each Teradata Enterprise stage, the following elements must be in effect:
򐂰 The name and data type of each database column corresponds to a DS
򐂰 Both DS and Teradata support null values, and a null value in a database
򐂰 DataStage gives the same name to its columns as the Teradata column
name. However, though DS column names can appear in either upper or
lower case, Teradata column names appear only in upper case.
13.7.5 Teradata Enterprise stage data type mapping

Teradata databases are not supported by the orcdbutil utility. It is important to
verify the correct Teradata to DS parallel data mapping, as shown in Table 13-9.

Teradata Data Type DS Parallel Data Type
byte(n) raw[n]
byteint int8
char(n) string[n]
date date
decimal[p,s] decimal[p,s] where p is the precision and S is the scale
double precision dfloat
float dfloat
graphic(n) raw[max=n]
integer int32
long varchar string[max=n]
long vargraphic raw[max=n]
numeric(p,s) decimal[p,s]
real Dfloat

Teradata Data Type DS Parallel Data Type
smallint int16
time time
timestamp timestamp
varbyte(n) raw[max=n]
varchar(n) string[max=n]
vargraphic(n) raw[max=n]
Important: Teradata data types that are not listed in the table cannot be used
in the Teradata Enterprise stage, and generate an error at runtime.
Aggregates and most arithmetic operators are not allowed in the SELECT clause
of a Teradata Enterprise stage.
13.7.6 Specifying Teradata passwords with special characters

Teradata permits passwords with special characters and symbols. To specify a
Teradata password that contains special characters, the password must be
surrounded by an “escaped” single quote as shown, where pa$$ is the example
password: \’pa$$\’
13.7.7 Teradata Enterprise settings

In the Teradata Enterprise stage, the DB Options property specifies the
connection string and connection properties in the following form:
user=username,password=password[,SessionsPerPlayer=nn][,RequestedSessio
ns=nn]
In this example, SesionsPerPlayer and RequestedSessions are optional

connection parameters that are required when accessing large Teradata
databases.
By default, RequestedSessions equals the maximum number of available

sessions on the Teradata instance, but this can be set to a value between 1 and
the database vprocs.

The SessionsPerPlayer option determines the number of connections each
DataStage parallel player opens to Teradata. Indirectly, this determines the
number of DataStage players, the number of UNIX processes, and the overall
system resource requirements of the DataStage job. SessionsPerPlayer must be
set in the following manner:
RequestedSessions = (sessions per player * the number of nodes *
players per node)
The default value for the SessionsPerPlayer suboption is 2.
Setting the SessionsPerPlayer too low on a large system can result in so many
players that the job fails due to insufficient resources. In that case
SessionsPerPlayer must be increased, and RequestedSessions must be
decreased.
13.7.8 Improving Teradata Enterprise performance

Setting the environment variable, $APT_TERA_64K_BUFFERS, might
significantly improve performance of Teradata Enterprise connections depending
on network configuration. By default, Teradata Enterprise stage uses 32 K
buffers. (Note that 64 K buffers must be enabled at the Teradata server level).
13.8 Netezza Enterprise stage

Starting with release 7.5.2, DataStage supports Netezza Performance Server
(NPS) targets on AIX®, Linux® Red Hat, Linux SuSE, and Solaris platforms
using the Netezza Enterprise stage. Netezza sources are supported through
ODBC.
Documentation for the Netezza Enterprise stage is installed with the DataStage
client, but is not referenced in the documentation bookshelf. You can find the
following installation and developer documentation for the Netezza Enterprise
stage in the Docs/ subdirectory in the installed DataStage client directory:
򐂰 Connectivity Reference Guide for Netezza Servers
򐂰 Netezza Interface Library

13.8.1 Netezza write methods
The Netezza Enterprise stage is a write stage. The stage takes bulk data from a
data source and writes that data to a specified destination table in NPS. You can
write data to NPS using two available load methods, as listed in Table 13-10.
Table 13-10 Load methods

Load Method Description Requirements
Netezza Load Uses NPS nzload utility to LOAD privileges for the target table; data in the source
load directly to target NPS database is consistent, contains no default values,
table single byte characters only, uses a predefined format
External Table Writes to an external table in If the data source contains default values for table
NPS; data is then streamed columns and uses variable format for data encoding
into the target table such as UTF-8
13.8.2 Limitations of Netezza Write stage

The following list details the limitations of Netezza Write stage:
򐂰 DataStage design column order and data types must match target NPS table.
During the write operation, the rows in the source database are mapped and
streamed into the destination database. The columns of the source database
are not mapped individually. You cannot interchange the order of the columns
when writing to the destination database. The Netezza Interface write
operator expects the column format in the destination database to be identical
to the schema of the source.
򐂰 Netezza Performance Server supports only ASCII table names and column
names. The following character sets are supported: v UTF-8 for NCHAR and
NVARCHAR data types v LATIN9 for CHAR and VARCHAR data types
򐂰 If the Netezza Enterprise stage encounters bad records during the write
operation, it aborts with an error. Bad input records are records that might be
corrupted in the source database. To ignore the corrupted records and
continue the write operation, you can specify the number of times to ignore
the corrupt records before the write operation stops. The Netezza Enterprise
stage ignores the bad records and continues to write data into NPS until the
number of bad input records equals the number specified.

13.8.3 Netezza Enterprise error logs
You can view the error logs to identify errors that occur during any database
operations, and view information about the success or failure of these operations.
By default the log files are created in the /tmp directory. When writing data to the
Netezza Performance Server by using the nzload method, the log files are
created in the /tmp directory on the client computer.
The following names are used for the log files:

򐂰 /tmp/database name.table name.nzlog
򐂰 /tmp/database name.table name.nzbad
When writing data to the Netezza Performance Server by using the External
Table method, the log files are created in the /tmp directory in the Netezza
Performance Server. The following names are used for the log files:
򐂰 /tmp/external table name.log
򐂰 /tmp/external table name.bad
Note: The log files are appended every time an error occurs during the write
operation.

14
Chapter 14. Connector stage guidelines

This chapter presents usage guidelines for Connector stages. As a general
guideline, new projects should give preference to Connector stages, and take
advantage of existing Enterprise equivalents in specific cases where these have
an edge in terms of performance.
The Connector library is constantly evolving, so the reader should always consult
the latest documentation and release notes.
This document is not intended to serve as an installation and configuration guide

for connectors, but instead focus on aspects related to data flow design.

14.1 Connectors and the connector framework
Until the advent of Connectors, DataStage (DS) relied on multiple types of stages
for database access. There were two major groups:
򐂰 API (DS Server-based)
򐂰 Enterprise (parallel) stages
The Connector library provides a common framework for accessing external data
sources in a reusable way across separate IS layers. The generic design of
Connectors makes them independent of the specifics of the runtime environment
in which they run.
The connectivity components are depicted in Figure 14-1.
Figure 14-1 Comparing types of connectivity components
Connectivity modules are used towards the following ends:

򐂰 Discover and import metadata
򐂰 Extract data
򐂰 Load data

Information Server goes beyond job execution engines such as DS Server and
parallel jobs, adding new tools such as Fast Track, Balanced Optimization and
Information Analyzer. Figure 14-1 on page 222 presents a comparison of the
places where each type can be used.
IS tools are implemented as J2EE applications running on top of WebSphere

Application Server. These tools also have the need to access external data
sources, discovering and importing metadata and loading/extracting data.
Plug-ins and Operators are components optimized for their corresponding

execution environments (the parallel framework naturally being the fastest).
Connectors implement a common framework that can be made use of in DS

Server and Parallel jobs, as well as Information Analyzer, FastTrack, and any
future tools, using the same runtime library.
They provide consistent user experience. They have a similar look and feel, with
minor GUI differences depending on the target type. The need for custom stage
editors is eliminated. There is a common stage editor and metadata importer with
rich capabilities.
There are a few important new concepts:

򐂰 LOB support
򐂰 Schema reconciliation
򐂰 Formalized metadata support
The connector framework is implemented in C++. However, the same runtime

libraries can be used in J2EE applications such as Information Analyzer and
FastTrack through JNI bridging. Design-time access to connectors is done
through the Connector Access Service.
The APIs and layers are not exposed to the general public as development APIs.
They are restricted to IBM, and enable the creation of new tools and
components. One example of a component built on top of the Connector API is
the Distributed Transaction stage (DTS), which supports the execution of
distributed transactions across heterogeneous systems through the XA Protocol.
Chapter 14. Connector stage guidelines 223

14.1.1 Connectors in parallel jobs
Figure 14-2 shows Connectors in the context of parallel jobs. PX Bridge is the
API layer that allows parallel operators to access the Connector runtime.
A similar bridge for DS Server jobs is named SE Bridge. The same Connector
runtime can be used in Java™ applications through a JNI layer.
Figure 14-2 Connectors in parallel jobs
The job is designed in the DataStage Designer client application, which runs on
Windows only. The job can be started and stopped from the DataStage Director
client application. The clients access ISF server through internal http port 9080
on the WebSphere Application Server. The stage types, data schemas, and job
definitions are saved to the common metadata repository.
When the job starts, the DataStage Enterprise engine starts a PX Bridge
Operator module for each Connector stage in the job. The PX Bridge acts as the
interface between the DataStage Enterprise framework and Connectors. The
Connectors are unaware of the type of the runtime environment.
The PX Bridge performs schema reconciliation on the server on which the job is
initiated. The schema reconciliation is the negotiation process in which the PX
Bridge serves as the mediator between the framework that offers the schema
definition provided by the job design (the schema defined on the Connector
stage’s link) and the Connector that offers the actual schema definition as
defined in the data resource to which it connects. During this process, the
attempt is made to agree on the schema definition to use in the job. The
differences that might be acceptable are as follows:
򐂰 Separate column data types for which data can be converted without data
loss
򐂰 Unused columns that can be dropped or ignored

The job might be configured to run on multiple execution nodes in parallel. The
DataStage Enterprise engine runs in distributed fashion. The communication and
synchronization among the running nodes is performed using the secure shell
and secure shell daemon mechanism. On each of the execution nodes
Connector instances are reading or writing portions of the data in parallel. This
provides for a high data throughput, especially when accessing resources that
support parallel and distributed data access, for example a DB2 database with a
database partitioning feature (DPF).
14.1.2 Large object (LOB) support

LOB data of any size can be moved from a source to target in a piecemeal
fashion. The target connector interacts with the source connector to read and
write the LOB fragments. This is depicted in Figure 14-3.
Instead of propagating the LOB itself (unnecessarily moving large amounts of

data across virtual links and operators all the way to the final target stage), the
job can propagate an LOB reference.
Figure 14-3 LOB Support in Connector stages

14.1.3 Reject Links
Reject functionality enhancements are as follows:
򐂰 Optionally append error code and error message data to the failing row data
򐂰 Conditionally filter which rows are rejected
򐂰 Ability to fail a job based on an absolute or percentage number of failures
The reject links in Connector stages is depicted in Figure 14-4. One major
implication of this reject functionality is the ability to output all records, regardless
of the presence of errors. This simplifies the development of real-time ISD jobs,
which need to synchronize the output from database stages with the InfoSphere
Information Services Director (ISD) Output stage. This is discussed in detail in
Chapter 16, “Real-time data flow design” on page 293.
Figure 14-4 Reject links in Connector stages

14.1.4 Schema reconciliation
Schema reconciliation is a mechanism to resolve discrepancies between
design-time and actual metadata. The Connector compares design-time and
propagated metadata with external metadata on a field-by-field basis. This is
Column Name Key SQL Type Length

OrderID Integer 4
CustomerID NChar 5
EmployeeID Integer 4 Compare Metadata
OrderDate Timestamp 23
RequiredDate Timestamp 23
Figure 14-5 Schema reconciliation
The goal is to reduce unpredictable problems that can stem from non-matching
metadata in the following ways:
򐂰 Identifying matching fields
򐂰 Performing data conversions for compatible data types
򐂰 Failing jobs that cannot be automatically corrected

14.1.5 Stage editor concepts
Each DS Server API and Enterprise stage has its own custom editor. For
Connectors, there is a Common Stage Editor that makes the design experience
much the same across all database connector types. The Connector Stage
Editor is depicted in Figure 14-6.
Figure 14-6 Connector Stage Editor
The Common Stage Editor for Connectors presents properties as a hierarchical,

contextual view. The properties, their list of valid values and their hierarchy are
defined by Connector definition files. There is one definition for each Connector
type. Definition describes the elements to be displayed. The Common Stage
Editor renders the presentation and drives the interaction with the user.

14.1.6 Connection objects
Connection objects encapsulate information to access external data sources.
Data such as user name, password and database name might be kept at a
central location and reused across multiple stages. This is depicted in
Figure 14-7.
Figure 14-7 Dragging connection objects
Connection objects might be dragged onto existing Connector stages, or onto the
canvas. When dragged into the canvas, a new Connector stage is created.

14.1.7 SQL Builder
Figure 14-8 presents a picture of the SQL Builder. The SQL Builder is accessible
by clicking an icon adjacent to the User-Defined SQL field.
Figure 14-8 The SQL builder

14.1.8 Metadata importation
Metadata imported through connectors creates shared metadata as well as
DataStage table definitions.
Table definitions are associated with the data connection used to import the
metadata. This is illustrated in Figure 14-9.
Figure 14-9 Metadata importation
Certain Connector stages take advantage of the Dynamic Metadata Interface

(DMDI). DMDI is an interface provided by the connector framework employed by
connectors to perform metadata importation to populate properties and column
definitions. The following stages use DMDI:
򐂰 Essbase
򐂰 Salesforce
򐂰 XML Pack
These connectors use DMDI to display a GUI to browse metadata objects, then
to import them and to set stage properties. The DMDI is launched from the stage
editor.

14.2 ODBC Connector
The ODBC Connector provides access to database resources through the
industry-standard ODBC interface. It is compliant with the ODBC standard
version 3.5, level 3. ODBC Connector accesses databases through the
user-defined ODBC data source name (DSN) definitions, which can be created
for a variety of ODBC drivers. On Windows, DSN definitions are created through
the built-in ODBC Driver Manager application and are stored in the Windows
registry. On UNIX systems, ODBC DSN definitions are provided in a special
ODBC initialization file pointed to by the ODBCINI environment variable.
The Connector supports three execution contexts for DataStage jobs:

򐂰 Source
򐂰 Target
򐂰 Lookup
When used in the source context, the Connector extracts data from the database
by executing SELECT SQL statements. It provides this data to other stages in
the job for transformation and load functions. When used in the target context,
the Connector loads, updates, and deletes data in the database by executing
INSERT, UPDATE, and DELETE SQL statements. The lookup context is similar
to the source context, with the difference that the SELECT SQL statements are
parameterized. The parameter values are provided dynamically on the input link
by the other stages in the job.
The ODBC Connector supports a special type of link called reject link. The
Connector can be configured to direct the data that it cannot process to the reject
link, from which it can be sent to any stage in the job (For example, the
Sequential File stage). The rejected data can later be inspected and
re-processed. Reject links provide for the option to continue running the job when
an error is detected, instead of interrupting the job at that moment. As shown in
Figure 14-10 on page 233, reject links are supported in lookup and target
context.

Figure 14-10 ODBC connector usage patterns
The Connector allows the option of passing LOBs by reference, rather than by
extracting the data and passing it inline into the job flow. When configured to
pass LOB values by reference, the ODBC Connector assembles a special block
of data, called a locator or reference, that it passes into the job dataflow. Other
Connectors are placed at the end of the job dataflow. When the LOB locator
arrives at the other Connector stage, the Connector framework initiates the
retrieval of the actual ODBC data represented by the reference and provides the
data to the target Connector so that it can be loaded into the represented
resource. This way it is possible to move LOB data from one data resource to

another where the size of the data is measured in megabytes and gigabytes,
without having to move the actual data through the job. The drawback is that the
data passed this way cannot be altered by the intermediate stages in the job.
The Connector supports array insert operations in target context. The Connector
buffers the specified number of input rows before inserting them to the database
in a single operation. This provides for better performance when inserting large
numbers of rows.
The Connector user uses the SQL Builder tool to design the SQL statements.
The SQL Builder tool is a graphical tool that enables construction of SQL
statements in a drag and drop fashion.
The Connector provides a mechanism for executing special sets of SQL

statements before processing the data. These statements are typically used to
initialize the database objects (for example to create a new table or truncate the
existing table before inserting data into the table).
The Connector supports metadata operations, such as the discovery of database

objects and describing these objects.
The IBM Information Server comes with a set of branded ODBC drivers that are
ready for use by the ODBC Connector. On Windows, the built-in driver manager
is used. On UNIX, a driver manager is included with the IBM Information Server
installation.
14.3 WebSphere MQ Connector

WebSphere MQ Connector provides access to message queues in the
WebSphere MQ enterprise messaging system. It provides the following types of
support:
򐂰 WebSphere MQ versions 5.3 and 6.0, and WebSphere Message Broker 6.0
for the publish/subscribe mode of work.
򐂰 MQ client and MQ server connection modes. The choice can be made
dynamically through the special connection property in the Connector’s stage
editor.
򐂰 Filtering of messages based on various combinations of message header
fields. Complex filtering conditions might be specified.
򐂰 Synchronous (request/reply) messaging. This scenario is configured by
defining both input link (for request messages) and output link (for reply
messages) for the Connector’s stage.

򐂰 Publish/Subscribe mode of work. Both the WebSphere MQ broker (with
MQRFH command messages) and WebSphere Message Broker (with
MQRFH2 command messages) are supported. The Connector stage can be
configured to run as a publisher or as a subscriber. Dynamic registration and
deregistration of publisher/subscriber is supported.
򐂰 MQ dynamic queues, name lists, transmission queues and shared cluster
queues for remote queue messaging
򐂰 Designated error queues and standard reject links.
The MQ Connector usage patterns are illustrated in Figure 14-11.
Figure 14-11 The MQConnector usage patterns

14.4 Teradata Connector
The Teradata Connector includes the following features of the ODBC Connector:
򐂰 Source, target and lookup context
򐂰 Reject links
򐂰 Passing LOBs by reference
򐂰 Arrays
򐂰 SQL Builder
򐂰 Pre/post run statements
򐂰 Metadata import
It supports Teradata server versions V2R6.1 and V2R6.2 and Teradata client
TTU versions V8.1 and V8.2. The Connector uses CLIv2 API for immediate
operations (SELECT, INSERT, UPDATE, DELETE) and Parallel Transporter
Direct API (formerly TEL-API) for bulk load and bulk extract operations.
Parallel bulk load is supported through LOAD, UPDATE, and STREAM operators
in Parallel Transporter. This corresponds to the functionality provided by the
FastLoad, MultiLoad, and TPump Teradata utilities, respectively. When the
UPDATE operator is used it supports the option for deleting rows of data
(MultiLoad delete task).
Parallel bulk export is supported through the EXPORT operator in Parallel

Transporter. This corresponds to the functionality provided by the FastExport
Teradata utility. The Connector persists the bulk-load progress state and provides
sophisticated support for restarting the failed bulk-load operations. The
Connector uses a designated database table for synchronization of distributed
Connector instances in the parallel bulk-load.
A limited support for stored procedures and macros is also available.
The Teradata usage patterns are illustrated in Figure 14-12 on page 237.

Figure 14-12 Teradata Connector usage patterns
14.4.1 Teradata Connector advantages

The following list details the Teradata Connector advantages:
򐂰 Parallel MultiLoad capability including MPP configurations
򐂰 Parallel immediate lookups and writes
򐂰 Array support for better performance of immediate writes
򐂰 Reject link support for lookups and writes on DS Enterprise Edition
򐂰 Reject link support for bulk loads
򐂰 Cursor lookups (lookups that return more than one row)
򐂰 Restart capability for parallel bulk loads
򐂰 MultiLoad delete task support
򐂰 Support for BLOB and CLOB data types
򐂰 Reject link support for missing UPDATE or DELETE rows
򐂰 Error message and row count feedback for immediate lookups/writes
򐂰 Parallel synchronization table

14.4.2 Parallel Synchronization Table
The following list details the Parallel Synchronization Table properties:
򐂰 Used for coordination of player processes in parallel mode
򐂰 Now optional, connector runs sequentially if not specified
򐂰 Can be used for logging of execution statistics
򐂰 Connector stage can use its own sync table or share it
򐂰 Primary key is SyncID, PartitionNo, StartTime
򐂰 Each player updates its own row, no lock contention
򐂰 Management properties for dropping, deleting rows
14.4.3 Parallel Transport operators

Table 14-1 shows a list of advantages and disadvantages of the Parallel
Transporter operators.
Table 14-1 Parallel Transporter operators

Operator Equivalent Advantages Disadvantages
Utility
Export FastExport Fastest export method. Uses utility slot, No single-AMP

SELECTs.
Load FastLoad Fastest load method. Uses utility slot, INSERT only,
Locks table, No views, No
secondary indexes.
Update MultiLoad INSERT, UPDATE, DELETE, Uses utility slot, Locks table, No
Views, Non-unique secondary unique secondary indexes, Table
indexes. inaccessible on abort.
Stream TPump INSERT, UPDATE, DELETE, Slower than UPDATE operator.

Views, Secondary indexes, No
utility slot, No table lock.
14.4.4 Cleanup after an aborted load or update

Take the following steps for cleanup after an aborted load or update operation.
1. DROP TABLE TableName_LOG;
2. DROP TABLE TableName_ET;
3. DROP TABLE TableName_UV;
4. DROP TABLE TableName_WT;
5. RELEASE MLOAD TableName; (Update operator only)

14.4.5 Environment variables for debugging job execution
Table 14-2 shows the environment variables for debugging job execution.
Table 14-2 Environment variables for debugging

CC_MSG_LEVEL=2 Turns on debugging output in the Director job log, displays
property values, generated SQL
CC_MSG_LEVEL=1 Full trace of the connector method calls
CC_TERA_DEBUG=1 Dump of the CLIv2 structures after each call to the

CC_MSG_LEVEL=2 Teradata client
CC_TERA_DEBUG=4 Turns on the Parallel Transporter’s trace output which is

written to a filename beginning with
“TD_TRACE_OUTPUT” in the project directory
14.4.6 Comparison with existing Teradata stages

The following list details comparisons with Teradata stages:
򐂰 Limited support for stored procedures and macros, but the Stored Procedure
plug-in is still better suited for it.
򐂰 No more utilities, named pipes, control scripts, or report files.
򐂰 Interface with the Parallel Transporter is through a direct call-level API.
򐂰 Error messages are reported in the DataStage Director log.
򐂰 MultiLoad plug-in jobs that use advanced custom script features cannot be
migrated to use the Teradata Connector.
򐂰 Number of players is determined by the PX Engine config file.

In Figure 14-13 we illustrate the Teradata stages and their relation to Teradata
Client APIs.
Teradata Teradata Teradata Teradata Teradata

DataStage MultiLoad Load API Enterprise Connector
plug-in plug-in plug-in
Stored
Procedure
plug-in
MultiLoad FastLoad CLIv2 Parallel Transporter
Teradata
TPump
client
FastExport
Figure 14-13 Teradata stages and their relation to Teradata client APIs
In Table 14-3, we present a feature comparison among all Teradata stage types.
Table 14-3 Teradata features for stage types

Feature / Teradata Teradata TDMLoad Teradata Load Teradata
Requirement Connector Operator Plug-in Plug-in API
Plug-in
Read, bulk Export YES Export only Export only No Read only
Write, bulk Load YES Load only Load only Load only Write only
Update/Upsert YES No YES No Yes
Sparse Lookup YES No No No YES
Array Support YES Limited Limited Limited support NO

support support (Fixed sizes:
(Fixed sizes: (Fixed sizes: 32/64 K)
32/64 K) 32/64 K)
Checkpoint support YES No Yes Yes No
EOW support YES No Yes No No
Reject Link support YES No No No No
Multiple-input-Links YES No No No No
DTS-nonXA support YES No No No No

New TD Data-Types: Supported Not supported Not Not supported Not
Decimal(38), BIGINT supported supported
Teradata basic client Required Required Required Required Required

(TTU)
Teradata PT client Required Not Required Not Required Not Required Not
(TPT) Required
stage Execution Both Both Sequential Sequential Sequential

Mode (Sequential /
Parallel)
Canvas (PX / Server) Both PX Both Both Both
In Table 14-4 we present a mapping between the Teradata Enterprise

environment variables and Connector functionality.
Table 14-4 Mapping of variables to functionality

Environment Variable Setting Description Equivalent in TDCC
$APT_TERA_SYNC_DATABASE [name] Starting with v7, specifies stage Property:

the database used for the “sync database”
terasync table.
$APT_TERA_SYNC_USER [user] Starting with v7, specifies stage Property:

the user that creates and “sync user”
writes to the terasync
table.
$APT_TER_SYNC_PASSWORD [password] Specifies the password for stage Property:

the user identified by “sync password”
$APT_TERA_SYNC_USER.
$APT_TERA_64K_BUFFERS Enables 64K buffer Automatic (checks

transfers (32K is the supported values
default). Might improve from DBS
performance depending on configuration and
network configuration. sets the buffer-size to
(32K/64K or 1M).
$APT_TERA_NO_ERR_CLEANUP This environment variable stage Property

is not recommended for (under “Bulk Access”
general use. When set, this category):
environment variable might “Cleanup Mode”
assist in job debugging by (with drop-down
preventing the removal of options: keep/drop).
error tables and partially
written target table.

$APT_TERA_NO_PERM_CHECKS Disables permission Default behavior
checking on Teradata
system tables that must be
readable during the
TeraData Enterprise load
process. This can be used
to improve the startup time
of the load.
14.5 DB2 Connector

The DB2 Connector includes the following features of the ODBC Connector:
򐂰 Source
򐂰 Target and lookup context
򐂰 Reject links
򐂰 Arrays
򐂰 SQL Builder
򐂰 Supports DB2 version V9.1
The Connector is based on the CLI client interface. It can connect to any
database cataloged on the DB2 client. The DB2 client must be collocated with
the Connector, but the actual database might be local or remote to the
Connector.
Separate the sets of connection properties for the job setup phase (conductor)
and execution phase (player nodes), so the same database might be cataloged
differently on conductor and player nodes.
The Connector provides support for the following tasks:

򐂰 Specifying DB2 instance dynamically (through connection properties), which
overrides the default environment settings (DB2INSTANCE environment
variable).
򐂰 XML data type in DB2 V9.1.
򐂰 DB2 DPF. A job with a DB2 Connector target stage might be configured to
assign on execution player node with each DB2 partition, and to write data to
the partitioned database in parallel, providing dramatic performance
improvement over sending the data to the same partition node and forcing
DB2 to redirect data to corresponding partitions.
򐂰 DB2 bulk load functionality. The invocation of bulk load is done through the
CLI interface. Parallel bulk load is also supported for DB2 with DPF.

In Figure 14-14 we illustrate the DB2 Connector usage patterns.
Figure 14-14 DB2 connector usage patterns

14.5.1 New features
In terms of functionality, the DB2 Connector offers more capabilities than all of
the existing stages. Specifically it provides support for the following elements:
򐂰 XML data type
򐂰 LOBs, with data passed either inline, or by a reference mechanism. The latter
allows for LOB data of any size to be moved from a source to a target.
򐂰 Client-server access to DB2 server. This overcomes a limitation with the EE
stage that can only be used in homogenous environments where DataStage
and DB2 are on identical (or the same) servers.
򐂰 Options to control bulk load than EE Operator.
򐂰 Design time capabilities, such as metadata import to the common model, and
enumeration of server-side information.
14.5.2 Using rejects with user-defined SQL

In this section we describe using rejects with user-defined SQL (UserSQL).
UserSQL without the reject link

The following list describes the use of UserSQL without the reject link:
򐂰 All statements in the UserSQL property are either passed for all of the input
records in the current batch (as specified by the Array size property) or none.
In other words events in previous statements do not control the number of
records passed to the statements that follow.
򐂰 If FailOnError=Yes, the first statement that fails causes the job to fail and the
current transaction, as depicted by the Record Count property, is rolled back.
No more statements from the UserSQL property are executed after that. For
example, if there are three statements in the property, and the second one
fails, it means that the first one has already executed, and the third one is not
executed. None of the statements have its work committed because of the
error.
򐂰 If FailOnError=No, all statements still get all the records but any statement
errors are ignored and statements continue to be executed. For example, if
there are three statements in the UserSQL property and the second one fails,
all three are executed and any successful rows are committed. The failed
rows are ignored.

UserSQL with the reject link
The following list describes the use of UserSQL with the reject link:
򐂰 All statements in the UserSQL property are either passed all of the input
records in the current batch (as specified by the Array size property) or none.
In other words, events in previous statements do not control the number of
records passed to the statements that follow.
򐂰 All the rows in each batch (as specified by the Array size property) are either
successfully consumed by all statements in the UserSQL property, or are
rejected as a whole. This is important to preserve the integrity of the records
processed by multiple statements that are expected to be atomic and
committed in the same transaction. In other words, the connector tries to
eliminate the possibility of having each statement successfully consume a set
of rows.
򐂰 If any of the rows in any statement in the UserSQL property are not
consumed successfully, the processing of the current batch is aborted and
the whole batch of records is sent to the reject link. The statements that follow
the failed statement are not executed with the current batch. The processing
resumes with the next batch.
򐂰 To preserve the consistency of records in each batch, the connector forces a
commit after every successful batch and forces a rollback after every failed
batch. This means the connector overrides the transactional behavior
specified by the Record Count property.
14.5.3 Using alternate conductor setting

When the connector is used on the parallel canvas, the engine spawns several
processes when the job runs. The main process is called the conductor. The
other processes are called players. These processes can run on the same
machine or on their own separate machines as defined in the PX configuration
file (default.apt). In the conductor process the connector performs metadata
discovery (describes the table that is later read or written to) and negotiates the
execution options with the engine. Later, in each player, the data is actually read
or written according to the setup performed in the conductor.

The following list details important facts and restrictions: in using alternate
conductor settings:
򐂰 There is always only one conductor process per job execution regardless of
the number of stages used in the job or the number of nodes defined in the
configuration file.
򐂰 Only one DB2 client library can be loaded in a process for the life of the
process. This means that the conductor process loads the client library
specified by the first DB2 stage in the job that it loads.
򐂰 Only one DB2 instance can be accessed from one process for the duration of
the process.
The following list describes common scenarios when the alternate conductor
setting might be needed:
򐂰 Scenario 1: All nodes run on the same physical machine
If there are DB2 client versions installed on the system or if there multiple
instances defined, it is important that all DB2 stages in the job specify the
same conductor connection parameters, otherwise the job fails because of
the aforementioned facts/restrictions. Using the alternate conductor setting
helps achieve this.
򐂰 Scenario 2: Each node runs on a separate machine
In this case the conductor might not run on the same machine as the player
processes. The same remote database that the connector is trying to access
might be cataloged differently on each node, and it might appear under a
different instance or database than it does to the players. Using the alternate
conductor setting allows you to use a separate instance/database for the
conductor process.
14.5.4 Comparison with existing DB2 stages

In determining the placement of the DB2 Connector in DataStage when
compared to the three other DB2 stages (DB2 EE Operator, DB2 API plug-in,
and DB2 Load plug-in) you must consider both functionality and performance.
In terms of functionality, the DB2 Connector offers more capabilities than all of
the existing stages, with new features as listed in 14.5.1, “New features” on
page 244. There are four DB2 stages available to DataStage users, as listed in

Table 14-5 DB2 stages available to users
Name Technology Where it Runs Capabilities
DB2 API Plug-in Server, Parallel Query, Upsert, Lookup
DB2 Load Plug-in Server, Parallel Load
DB2 EE Operator Parallel Query, Upsert, Lookup, Load
DB2 Connector Connector Server, Parallel Query, Upsert, Lookup, Load
The plug-in stages collectively offer approximately the same functionality as both
the operator and connector. However, on the parallel canvas the two plug-in
stages perform significantly worse than the operator and connector. Because
these stages offer poor performance and do not provide any functionality that is
not offered by the connector or operator, the emphasis of this document shall be
on a comparison of the DB2 Connector and DB2 EE Operator.
In terms of performance, the DB2 Connector performs significantly better than

the DB2 API and DB2 Load plug-ins (between three and six times the throughput
of these stages). When compared to the EE stage, the connector performs better
for INSERT. SELECT performance of the DB2 Connector can vary when
compared to the EE stage, depending on the number and type of columns, and
the number of parallel nodes.
The DB2 Connector must be considered the stage of choice in almost every
case. Because it offers more functionality and significantly better performance
than the DB2 API and DB2 Load stages, the connector is the choice, rather than
these stages. The DB2 Connector must be used instead of the EE stage in most
cases.
The EE stage has an advantage in the following cases:

򐂰 Bulk load performance
򐂰 Select performance, in certain situations, especially with a large number of
columns
DB2 Enterprise configuration is significantly more complex than the configuration

of the DB2 Connector. If obtaining absolutely the best performance is paramount,
and the added configuration complexity is acceptable, then DB2 Enterprise must
be used.

Architectural differences
The primary difference between the connector and EE stage is that the former
uses the CLI interface for all of its interactions to enable client-server
communications for all functions. By contrast, the EE stage only uses a
client-server interface on the conductor node for querying table definitions and
partitioning information. The player nodes have to run on the DB2 server. The EE
stage can only be used with a partitioned DB2 database. That is, the DB2
Database Partitioning Feature (DPF) must be enabled.
The deployment of the EE stage is illustrated in Figure 14-15.
Conductor Node Player Node 1

DB2 EE
Stage CLI DB2
DB2 EE
Part 1
Stage
Player Node N
DB2
DB2 EE
Part N
Stage
Figure 14-15 DB2 enterprise deployment architecture
In the case of the connector, the player nodes might correspond to DB2 partitions
(as is the requirement for the EE stage), but this is not a requirement. Because
the connector communicates with DB2 through the remote client interface, the
deployment options are much more flexible. This is depicted in Figure 14-16.
Conductor Node Player Node 1 DB2 Server

DB2 DB2
Connector CLI DB2 DB2
Connector CLI
DB2
Player Node N
DB2 DB2
Connector CLI
Figure 14-16 DB2 connector deployment architecture

There are advantages and disadvantages to both approaches:
򐂰 The EE stage requires that the DataStage engine and libraries be installed on
the DB2 nodes. This can be achieved by running a script that is provided as
part of the DataStage install. This script copies only PX core libraries, but
does not copy any of the operators, including DB2 EE stage. These additional
libraries have to manually copied to the DB2 partitions.
Advantage: connector
򐂰 The EE stage requires a homogeneous environment. The DataStage server
nodes and DB2 server nodes must be running the same hardware with the
same operating system. The connector has no such requirement. This means
there are customer situations where the EE stage cannot be used.
Advantage: connector
򐂰 The EE stage maintains a direct local connection between the operator and
the DB2 server (because it is always co-resident). This might offer
performance advantages compared to the connector, which always
communicates to the DB2 server through TCIP/IP, even when it is installed
locally to the DB2 server.
Advantage: operator

14.6 Oracle Connector
The Oracle Connector includes these features of the ODBC Connector:
򐂰 Source
򐂰 Target and lookup context
򐂰 Reject links
򐂰 Arrays
򐂰 SQL Builder
In addition, it supports bulk loads and Oracle partitioning.
The Connector works with Oracle versions 10g and 11g. It supports connecting
to an Oracle database through Oracle Full Client or Oracle Instant Client (Basic
or Basic Lite). Oracle Connector usage patterns are depicted in Figure 14-17.
Figure 14-17 Oracle connector usage patterns

14.6.1 New features and improvements
The following list details improvements made to the Oracle Connector:
򐂰 Distributed transactions
– Support for guaranteed delivery of transactions arriving in form of MQ
messages. In case of success, the messages are processed by the job
and the data is written to the target Oracle database. In case of a failure
the messages are rolled back to the queue.
– To use the Distributed Transaction stage, you need MQ 6.0 and Oracle
10g R2.
򐂰 Built-in Oracle scalar data types are supported, including BLOB, CLOB,
NCLOB, BFILE, LONG, and LONG RAW data types.
򐂰 XMLType columns and object tables are supported.
򐂰 PL/SQL anonymous blocks with bind parameters are supported.
򐂰 Support for a rich set of options for configuring reject links. Reject links are
also supported in bulk load mode
򐂰 Pre- and Post- SQL operations are supported:
– SQL statement, multiple SQL statements or PL/SQL anonymous block
might be specified.
– The statements might be configured to run at the job or at the node level.
– The statements might be specified in the stage UI or in external files.
򐂰 Rich metadata import functionality:
– Table selection based on table type (table, view, IOT, materialized view,
external table, synonym), the table owner or the table name pattern.
– Supports importing PK, FK, and index information.
򐂰 Table action:
– Performing Create, Replace, or Truncate table actions in the job is
supported before writing data to the table.
– Input link column definitions automatically used to define target table
columns.
򐂰 Fast read, write, and bulk load operations:
Internally conducted testing showed performance improvements over Oracle
Enterprise stage.
򐂰 Rich set of options for configuring parallel read and write operations.

򐂰 Provides control over the Transparent Application Failover (TAF) mechanism
in environments such as Oracle RAC.
Includes storing TAF notifications in the job log to inform the user about the
failover progress.
򐂰 Oracle or OS authentication is supported for the connector at job runtime.
14.6.2 Comparison with Oracle Enterprise

Table 14-6 shows a mapping between Oracle-related environment variables used
by existing stages to the Oracle connector. Behavior of Oracle Connectors is no
longer affected by any environment variables, with the exception of the Oracle
client vars such as ORACLE_HOME and ORACLE_SID.
Table 14-6 Equivalence of Oracle-related environment variables

Existing Oracle-Related Env Vars Connector Equivalent
$ORACLE_HOME and $ORACLE_SID These are Oracle environment variables. They apply to the
connector the same way they apply to the Enterprise stage.
They are used by the Oracle client library to resolve Oracle
service name to which to connect.
$APT_ORAUPSERT_COMMIT_ROW_I No environment variable. The connector has a record count

NTERVAL property that can be used to control this.
$APT_ORAUPSERT_COMMIT_TIME_I No environment variable and no property.

NTERVAL
$APT_ORACLE_LOAD_OPTIONS Not applicable to the connector. The connector does not use
SQL*Loader for bulk-load hence the loader control and data
files are not applicable to the connector. The connector uses
OCI Direct Path Load API. The settings to control the load
process are provided through connector properties.
$APT_ORACLE_LOAD_DELIMITED None, because the connector does not use SQL*Loader.
$APT_ORA_IGNORE_CONFIG_FILE_ None, because the connector does not use SQL*Loader.

PARALLELISM
$APT_ORA_WRITE_FILES None. The connector does not use SQL*Loader for bulk-load
operations.
DS_ENABLE_RESERVED_CHAR_CO None. The connector relies on the framework to take care of

NVERT this. The connector itself supports importing, reading or
writing to Oracle tables and columns that contain # and $
characters. No extra setting is needed for this.

14.7 DT stage
The Connector framework provides support for distributed two-phase XA
transactions in DataStage Enterprise jobs. The transactional data is carried by
MQ messages that arrive at the source queue. Each message might encompass
multiple database operations, and multiple messages might be grouped in a
single physical transaction.
The MQ Connector stage provides properties for the configuration of transaction

boundaries. It is possible to specify the number of source messages included in
each transaction or the time interval for collecting source messages in each
transaction. The MQ Connector uses a specially designated work queue as
temporary buffer storage for source messages that participate in transactions.
The retrieved messages are processed by any number and combination of

transformation stages, chosen from a rich palette of stage types provided by
DataStage. The processed messages result in rows of data that arrive on the
Distributed Transaction (DT) stage on one or more input links. Each input link on
the DT stage is associated with one external resource. The rows on each link are
sent to the designated resource (as insert, update, or delete operations on the
resource).
The DT stage reads messages from the work queue and updates external
resources with the data rows corresponding to those work queue message. The
reading of messages and writing to external resources is done in atomic
distributed transactions using the two-phase XA protocol.
The DT stage can also be used outside the context of a queuing application.
Transactions are committed upon EOW markers. This has application in ISD jobs
that need to guarantee transactional consistency, but do not involve the
processing of messages.

The DT stage is depicted in Figure 14-18 and described in detail in Chapter 16,
“Real-time data flow design” on page 293.
MQ DT
Connector Stage
Stage
Transformation
Stages
Transaction
Messages Resource
C
Resource
B
Source Work Resource
Queue Queue A
Figure 14-18 The DT stage

14.8 SalesForce Connector
The SalesForce Connector, depicted in Figure 14-19, supports load, extraction,
and delta extraction. It generates SQL statements based on user selection. Load
mode supports create and updates, as well as reject links. The connector uses a
Web service interface through an Axis library, and is implemented as a Java
connector using DMDI.
Figure 14-19 SalesForce connector

14.9 Essbase connector
The Essbase connector supports the extraction, delta extraction, and load of
data to and from Essbase databases. It performs hierarchical to relational
mapping of cube data. The connector supports parallel read and writes.
It is implemented in C++ using a 3rd party interface library from a partner, and
uses DMDI to allow selection from cube data. See Figure 14-20.
Select the link or the

Connector to edit
Figure 14-20 The Essbase Connector

14.10 SWG Connector
The SWG Connector, depicted in Figure 14-21, provides a bridge to SWG
Adapters. The first release is to support SAP. The SWGA Connector is
Adapter-neutral. Any adapter should plug in without a need to change the
connector. A DMDI implementation provides the means to configure the stage
and import metadata.
DataStage PX DataStage SE
PX Bridge Operator SE Bridge Plugin
Connector Framework
C++ JNI Layer

Connector
Java SWGA
Connector Connector
SWG
Adapters Adapter
plugin here:
Figure 14-21 SWG Connector

15
Chapter 15. Batch data flow design

Batch applications are always constrained by execution time windows. They
must complete the extraction, transformation, and load of large amounts of data
in limited time windows. If the execution fits in a certain expected amount of time,
performance requirements are met. However, it is not enough to meet initial
expectations narrowly. Jobs must be able to scale, as data volumes grow.
Parallel batch jobs must be implemented with parallel techniques that yield the
necessary scalability, which is the focus of this chapter. The basic assumption
throughout this discussion is that no applications other than DataStage (DS)
update the target databases during batch windows.
This chapter does not focus on individual stage parameters, but rather on data
flow design patterns. Individual stage types are the subject of other chapters in
this document. For instance, when dealing with database interfaces, there are
several options for each database type, along with their tuning parameters.

15.1 High performance batch data flow design goals
In this section we describe the main goals a DS application developer must keep
in mind when designing batch applications.
15.1.1 Minimize time required to complete batch processing

The first and foremost goal is to complete work in the available time, be it a daily,
weekly, or monthly process.
15.1.2 Build scalable jobs

It is not enough to meet initial expectations. As more data is processed, jobs
should complete in approximately the same amount of time as more hardware is
added (processors, memory, disks). Conversely, efficient designs make it
possible to reduce the execution time for the same amount of data as more
hardware resources are added. There are things that money can buy, such as
more hardware. However, bad job designs do not scale even with more
hardware, in most circumstances.
15.1.3 Minimize the impact of startup time

Jobs have a startup up time, which is the time spent by a job during its
preparation phase. During this phase the conductor process parses the OSH
script (the script created by the compilation), builds a job score (similar to a query
execution plan in database terms) and spawns child processes such as section
leaders and players. These are steps that cannot be parallelized. They are run
sequentially by the conductor process and they take a certain amount of time.
In most cases, the startup times tend to be small but they grow as the jobs grow
in size. They can take considerable time for a large number of jobs.
Jobs must be designed in a way that allows them to process large amounts of
data as part of a single run. This means input files must be accumulated and
submitted as input to a single job run. This way, instead of starting up each and
every job in a long sequence several times (such as once for each input file), the
sequences and their jobs are started up only once.
Once up and running, well-designed jobs are able to cope with large amounts of
data. The use of parallel techniques leads to more streamlined and efficient
designs that help mitigate the impact of startup times.

15.1.4 Optimize network, I/O and memory usage
The following list details the hardware resources in descending order of their cost
(this order assumes storage subsystems comprised of parallel disks):
1. Network
2. Disk
3. Memory
4. CPU
Network and disk I/O are the two slowest hardware resources in most systems.
Therefore, they must be used judiciously. For small amounts of data, they are not
a concern. But when there is a need to meet the demands of large environments,
these two resources are no longer negligible. There are a few aspects to keep in
mind that have impact on one or more of the resources listed:
򐂰 Minimize the landing of data to disk.
– Instead of writing and reading data to and from disk after each and every
small job, create jobs that do more processing as part of the same
execution.
– This was the typical approach for DS Server applications, which must be
avoided in the parallel world.
򐂰 Minimize interaction with the database Server.
– Extract/Load data in bulk.
– Avoid sparse lookups.
– Avoid extracting unnecessary data for lookups
For instance, avoid full table extracts when the number of input rows is
significantly smaller. One does not need to extract 100 million rows from a
database as a lookup table to process only a few thousand incoming
records.
򐂰 Avoid unnecessary re-partitioning.
Re-partitioning cost is not pronounced in SMP systems, but it adds to private
network usage in DataStage grid environments.
򐂰 Limit the amount of reference data for normal lookups.
– Single large normal lookup reference tables can consume a huge amount
of memory.
– Avoid full table extracts when the number of rows being processed is
significantly smaller.
– If the number of rows cannot be restricted, use a sorted merge instead of a
normal lookup.
Chapter 15. Batch data flow design 261

򐂰 Implement Efficient Transformations
Must implement efficient expressions and logic after data is in the job’s
address space. This has a direct impact on CPU usage.
For instance, instead of using a long “IF-THEN-ELSE IF-…” sequence in a
Transformer for conversions, use a small normal lookup to implement the
same logic.
15.1.5 Plan job concurrency and degrees of parallelism

Once application coding is finished and ready for production, the applications
must be able to be executed with varying degrees of parallelism, depending on
their complexity and amount of data to be processed:
򐂰 Contain the number of concurrent jobs.
– The goal is to avoid swamping the hardware resources with too many
concurrent jobs.
– Job sequencing and scheduling must be planned carefully.
򐂰 Maximize the degree of parallelism for larger, complex jobs.
򐂰 Reduce degree of parallelism for smaller, simpler jobs.
15.2 Common bad patterns

Common bad design patterns are a result of experience with the existing DS
Server technology. Other bad design patterns result from inexperience in high
volume batch processing.
Without going in to detail about the origins of bad practices, the following
sections discuss a few recurring patterns that unfortunately are common in the
field, and lead to less than optimal results to say the least. Bad practices
includes, but are not limited to, the topics discussed in the following subsections.
15.2.1 DS server mentality for parallel jobs

DS server and parallel jobs are similar from a Graphical User Interface
perspective providing the same basic paradigm: stages and links. However, they
implement fundamentally different execution frameworks.
DS Server provides a limited degree of intra-job concurrency: an active stage

and its surrounding stages are mapped to processes, and the number of active
stages determines the number of processes that are executed as part of a

Server job run. There is no support for partitioning parallelism, unless done
manually with multiple invocations of the same job.
The active/passive nature of DS Server job stages limits the types of stages and
prevents a data flow execution model. This ultimately prevents joins and merges
from being made available in DS Server. As a result, DS Server jobs follow
design patterns that include the following attributes:
򐂰 Singleton lookups.
A SQL statement is executed for each and every incoming row.
򐂰 Small jobs with frequent landing of data to disk.
The main reason being checkpoint and job restartability.
򐂰 All correlation must be done using hash files and Transformers.
Results from lack of join/merge functionality.
򐂰 Manual partitioning with multiple instances of a same job.
Unfortunately there is a tendency for seasoned DS Server professionals to adopt

the same techniques for parallel jobs. This must be avoided. Developers must
put effort into learning new techniques to make the most of the parallel
framework.
15.2.2 Database sparse lookups

A database sparse lookup is one of the most inefficient constructs in parallel
jobs. That is because it involves the invocation of an SQL statement for each and
every row flowing through the lookup stage.
This invocation involves preparing and sending one or more packets to the
database server, the database server must execute the statement with the given
parameters and results are returned to the lookup stage.
Even if the database server is collocated with DataStage, it is still a bad solution.
There is at least one round-trip communication between the Lookup stage and
the database server, as well as a minimum set of instructions on both sides
(DataStage and database) that is executed for each and every row.
By using bulk database loads and unloads, larger sets of data are transferred
between DataStage and the database. The usage of the network transports as
well as the code paths on each side are optimized. The best solution is then
achieved, at least in terms of DataStage interfacing.

This has implications in the way DataStage jobs are designed. Parallel jobs must
make extensive use of the following elements:
򐂰 Bulk database stage types;
򐂰 Other lookup techniques, such as joins and normal lookups.
In rare occasions, sparse lookups might be tolerated. However, that is only when
it is guaranteed that the number of incoming rows, for which the Lookup must be
performed, is limited to at most a few thousand records.
15.2.3 Processing full source database refreshes

There are cases when source systems (such as mainframes) are unable to
restrict the data transferred to data warehouses on a daily basis.
Those systems are not equipped with the right tools to only extract the net
difference since the last refresh. As a result, incremental transfers are not
possible and the entire content of the source database must be copied to the
target environment in each processing cycle (frequently on a daily basis).
That puts a tremendous amount of load across the board:

򐂰 Processor, memory and disk:
On the source system and especially on the target DataStage and DW
servers;
򐂰 Network
Networks are limited to a certain bandwidth and unless the infrastructure is
upgraded (assuming there is faster technology available), they just cannot go
any faster.
Source systems must implement a form of change data capture to avoid this
unnecessary burden.
15.2.4 Extracting much and using little (reference datasets)

Beware of extracting huge reference datasets but only using a small fraction of
them. If an input dataset to a batch process is significantly smaller than the target
database table, it is a poor use of time and resources to download the entire
table contents to a lookup file set or dataset. It is a waste of network, disk,
memory and processor resources. Unfortunately, this is a common, recurring
pattern.
The size of reference datasets and lookup file sets must be restricted using the
techniques presented in this chapter.

15.2.5 Reference data is too large to fit into physical memory
Beware of insisting on normal lookups when the reference data is too large to fit
into physical memory. There are frequent cases when, although it is clear the
reference data consumes all memory, developers insist on using normal lookups
instead of resorting to joins.
It might be the case of not only a single lookup, but the combination of all normal
lookups across multiple concurrent jobs that eat up all available memory and
leave little or no space in physical memory for other processes and applications.
Even worse, projects continue to consume not only the physical memory, but all
swap space as well. In these severe cases, DataStage throws fatal exceptions
related to fork() errors. The application must be changed to use the right
mechanisms:
򐂰 Restrict the size of the reference data
򐂰 Use sorted joins
15.2.6 Loading and re-extracting the same data

Instead of caching rows in the form of persistent datasets or file sets,
applications load records into a target database and re-extract the same
information over and over again. This is a bad design practice.
The place to store data that might be re-referenced in the same batch cycle is in
persistent datasets or file sets. As jobs process incoming data, looking up data,
resolving references and applying transformations, the data must be kept local to
DataStage. Only at the last phase of the Extract, Transform, and Load (ETL)
process can data be loaded to the target database by provisioning jobs.
15.2.7 One sequence run per input/output file

Another common pattern is the re-execution of entire job sequences for each and
every input or output file. There are cases when the application generates large
numbers of output files. But the file names are derived from values set as job and
sequence parameters. In this case, the entire sequence must be executed for
each and every output file.
Data values cannot be passed as job parameters. The application must be

structured in a way that it can process large amounts of data in a single job or
sequence run. The application is then able to scale across multiple processors
(and blades in a grid environment) by increasing the degree of parallelism in the
APT_CONFIG_FILE and by adding more hardware resources.

There are cases when having multiple job invocations is unavoidable, such as
when input files are received at separate times during the execution time window.
For example, files from a mainframe might be transferred to the ETL environment
at separate times during the 12:00am to 6:00am window. When this happens, it
is not a good option to wait until 6:00am to have all files ready and then start the
entire batch sequence. However, it is important to accumulate input files to an
extent, to limit the total number of sequence executions.
15.3 Optimal number of stages per job

It is a good practice to combine more logic inside a single job. The main purpose
in combining more logic inside a single job is to gain the following advantages:
򐂰 Avoid small jobs
򐂰 Limit the frequency in which data is landed to disk
We understand that combining too much logic inside a single job might saturate
the hardware with an excessive number of processors and used memory. That is
why it is important to know how much memory and processing power is
available. However, considering the difference between memory and disk access
speeds, we tend to favor larger jobs.
The number of stages for each job must be limited by the following factors:
򐂰 The amount of hardware resources available: processors and memory
򐂰 The amount of data in the form of normal lookup tables
– One or more large lookup tables might end up consuming most of
available memory.
– Excessively large normal lookups must be replaced by joins.
– The amount of data on the reference link must be restricted to what is
minimally necessary.
򐂰 Natural logic boundaries
– Such as when producing a reusable dataset or lookup fileset. The creation
of such reusable set must be implemented as a separate job.
– Database extraction and provisioning jobs must be isolated to separate
jobs.
– There are times when an entire dataset must be written out to disk before
the next step can proceed.
This is the case, for instance, when we need to upload a set of natural
keys to extract relevant reference data (see 15.6.1, “Restricting incoming
data from the source” on page 270).

There is not a one-size-fits-all. If there is enough hardware, it is better to do
perform more tasks in a single job:
򐂰 Let the framework orchestrate the execution of components in the parallel
flow
򐂰 It might require a well-tuned buffer space (scratch/buffer pools) possibly on
local disks or SAN (with multiple physically independent spindles)
This is the case for jobs that contain fork/join patterns.
15.4 Checkpoint/Restart
The parallel framework does not support checkpoints inside jobs. The
checkpoint/restart unit is the job.
DataStage sequences must be designed in a way they can catch exceptions and
avoid re-executing jobs already completed. DataStage sequences provide for the
ability to design restart-able sequences. The designer, however, must explicitly
use constructs to enable this behavior.
The DataStage Server approach has always been to land data frequently, after
small steps, as pointed out in previous sections. It is not the same with parallel
jobs. Again, the question is: where is the biggest cost? The answer is, in
descending order:
1. Network access
2. Disk
3. Memory
4. Processor
As a result, it is better to restart a bigger scalable, optimized job than to land data
after each and every small step.
15.5 Balanced optimization

In this section we discuss when it is best to use a database or DataStage for
transformations. DataStage and databases work in conjunction to achieve a
certain goal: loading clean, consistent data into useful warehouses.
There are certain tasks that ETL tools do better, and tasks at which databases
excel. It is not only a matter of capabilities, but also how many resources are
available to each, and their financial, operational, and licensing costs.
As stated before, network and database interfacing have the highest costs (That
is, loading data into and extracting from databases are costly operations).

15.5.1 Transformations inside the database
The following considerations are for performing transformations inside the DB.
The following list details the conditions that, if true, indicate that you should
consider performing transformations inside the DB:
򐂰 The database server is massively parallel and scalable
򐂰 There is enough HW capacity and licensing on the database side for the extra
processing cost for transformations
򐂰 All the data is contained inside the DB
򐂰 Data is already cleansed
򐂰 There are guarantees that set-oriented SQL statements does not fail
򐂰 For example, you are certain that a “SELECT .. INTO…” always runs
successfully.
򐂰 The database is one of the database types supported by the Balanced
Optimizer
If you decide to perform the transformations inside the DB, then you can
delegate transformation logic to the database server. That is what the Balanced
Optimizer (Teradata) was built for (and is a new component of the InfoSphere
product suite in Information Server 8.1). There is no need to extract and load
data back into the database because it is already there.
The key is the use of Balanced Optimizer, so the logic is designed inside the DS
paradigm, keeping the ability to draw data lineage reports. The use of pure SQL
statements defeats the purpose, as there is no support for data lineage and the
application becomes increasingly complex to maintain.
15.5.2 Transformations with DataStage

These are circumstances that favor performing transformations with DataStage:
򐂰 DataStage can scale in a separate environment, meaning:
– It should result in lower cost
– A grid environment might be set up so it scales way beyond the database
capabilities
– It does not interfere with complex database queries

򐂰 When there are limitations on the database side:
– The database does not scale well or becomes saturated with extra
transformation load
• Not a truly parallel database
• Not enough HW resources
– The database is not supported by the InfoSphere Balanced Optimizer
򐂰 Application needs functionality not available inside DBs, such as:
– QualityStage
– Custom plug-ins
򐂰 Must combine/correlate data from other sources
– Data not completely clean yet
Must be able to reject rows on a row-by-row basis
– Date from existing systems (SAP, other database flavors, non-relational
sources)
15.6 Batch data flow patterns

In this section we present design patterns that address important aspects that
have major impact in high performance batch applications. Some might say the
proposed patterns are overkill, they are too complicated, and the results do not
really pay off. Our experience demonstrates the opposite. Advice presented here
is based on a solid track record. Projects that follow these recommendations and
patterns are typically successful, and projects that do not follow them fail.
There is a multitude of topics covered in other chapters that are all still valid and
relevant. For example, consider the type of database stage for extracting and
loading data from and to databases. You need to use bulk database stages, or
loads and unloads take too long. This is dependent upon how efficiently DS
interoperates with target database systems.
The specifics on databases and other stage types and their properties are
described in the other chapters of this IBM Redbooks publication.

In terms of design patterns (the focus of this chapter) the four considerations that
have the most impact in high performance batch applications are as follows:
򐂰 Extracting reference data from source/target databases efficiently
򐂰 Correlating various datasets efficiently, such as incoming files and reference
data (once this reference data is already in DataStage format)
– Matching
– Joins, Lookups
򐂰 Transforming efficiently based on business rules (transformation expressions,
data quality
– Custom business rules
– Field mappings
– Standardization, survivorship
򐂰 Loading data into the target databases efficiently.
We mentioned in the list the subject of data quality (Matching, Standardization,

Survivorship). In this document we focus on efficient ways to bring candidate
records into DataStage. QualityStage practices are a subject of a separate CEDI
document and are not discussed here.
15.6.1 Restricting incoming data from the source

The first step in the process of implementing highly efficient batch processes is to
restrict the amount of data sent from a source system (for example, a mainframe)
to DataStage. The data must be restricted at the source so that only
modifications to data are transmitted and not the entire contents of the source
database.
If the entire source database is sent to DS for processing, it is an inefficient use

of network and processing resources. Processing windows are unnecessarily
long and the network proves to be the most significant bottleneck.
Networks have a certain bandwidth and do not yield a higher throughput unless
the entire network infrastructure is upgraded (assuming there is a faster network
technology available).
This means using Change Data Capture on the mainframe, for instance, and not
resending the entire source database. If it is not possible to restrict the source
data sent to DataStage, the application has to resort to full target table extracts.
Under those circumstances, you must resort to sorted joins instead of normal
lookups, as there is a high probability that the reference data does not fit in
memory.

15.6.2 A fundamental problem: Reference lookup resolution
The fundamental problem that pervades any type of database batch load is how
to correlate the input data to database rows in the most efficient way. This
includes the following common tasks:
򐂰 Finding out, for a given input record, if a corresponding record already exists
in the target DB
This helps determine if a new or existing SK must be created, or if an existing
record needs to be updated.
This applies to any Slowly Changing Dimension (SCD) Type (1 or 2), or any
relational database load scenario.
򐂰 Fetching candidate records for de-duplication/matching with QualityStage
By reference lookup we mean the process of looking up any type of information

(either fact or dimension tables, or any other type of table, for that matter) in the
target database, on behalf of input records.
There are two possible ways of accessing reference data:

򐂰 Export the entire contents of the target tables to datasets or file sets;
򐂰 Use database sparse lookups (that is, one singleton select for each incoming
row).
The first solution is inefficient if the size of the input data is small when compared
to the database table size. The second solution involves a method that must be
used for really small input files.
In case of SCD type 2, doing a full extract implies restricting the set to rows with
Current Flag set to 1, or a filter on Expiration Date. Even restricting the extracted
set to the most current records might represent an amount of data that is too
large to hold in memory, or that contains way more records than are needed for a
given processing cycle. Instead, we propose an approach that is based on the
optimization of the interaction between DataStage and the target database,
which involves the following elements:
򐂰 Restriction of the reference dataset to only what is minimally relevant for a
given run.
Avoid extracting reference data that is not referenced in the input files.
򐂰 Use of the optimal correlation strategy inside DataStage.
Use elements such as the join, normal lookup, change capture, and SCD.
򐂰 Use data already stored locally to DataStage in the form of datasets or
filesets.
Avoid repeated and redundant database extracts.

15.6.3 A sample database model
Figure 15-1 depicts an example that is used throughout this section. It consists of
three tables that are loaded from three separate input files.
Each input file maps directly to a table. The Source System Key (SSK) fields are
marked as *_SSK. Surrogate key (SK) fields follow the pattern *_SK.
DB Model
Customers Sales Products

Sale_SK
Cust_SK Prod_SK
Sale_SSK
Cust_SSK Prod_SSK
Cust_Name Cust_SK (FK)
Prod_Name
Cust_Address Prod_SK (FK)
Prod_Descr
Price
Cust_Src_TS Prod_Date
Qty
Cust_TS Prod_Src_TS
Date
Prod_TS
Sale_Src_TS
Sale_TS
Batch Load
Sale_SSK
Cust_SSK Prod_SSK
Cust_SSK
Cust_Name Prod_Name
Cust_Address Prod_SSK
Prod_Descr
Cust_Src_TS Price
Prod_Date
Qty
Prod_Src_TS
Date
Sale_Src_TS
Input Flat Files

Figure 15-1 Sample database model and input files
Input files contain a Source Timestamp (*_Src_TS). Database records carry

those values into the target tables. However, each record is assigned a
timestamp for the time the record was created or updated by the batch process.
In this chapter we abstract the type of slowly changing dimension. However, the
techniques described in the following sections apply to any SCD types.

15.6.4 Restricting the reference lookup dataset
The following process must be following when the number of input records is
significantly smaller than the number of records in the target table. It consists of
three basic steps:
1. Collect Unique SSKs across all input files.
2. Load those unique SSKs into a temporary table in the target database.
3. Extract the relevant set of reference records with a database join between the
target and temporary tables.
Collecting unique SSKs for same entity from multiple input

files
Follow these steps to collect unique SSKs for the same entity from multiple files:
1. Isolate unique sets of incoming SSKs. These unique SSK datasets help
restrict the set of relevant reference records for a given run.
As shown in Figure 15-2, each type of SSK is present in more than one file.
For instance, the customer SSK (Cust_SSK) is present in the Customers and
Sales flat files. All Cust_SSK values must be mapped to valid surrogate keys
Cust_SK. Instead of extracting reference data for each input file separately,
combine all the SSK types into separate datasets, one dataset per SSK type.
Figure 15-2 Collecting unique customer SSKs

SSKs are funneled and duplicates removed. The inputs to the sort stages are
hash partitioned by Cust_SSK. We recommend the use of explicit sort stages.
2. Load the unique SSKs into a temporary table as in Figure 15-3.
Figure 15-3 Loading Cust_SSKs into a temp table
Another pair of jobs, similar to the ones depicted in Figure 15-2 on page 273
and Figure 15-3 must be implemented for Product SSKs.
The net result of this phase is that each temp table (one for each SSK type)
contains all the unique SSKs for that type across all input files.
QS matching
QualityStage matching requires candidate records based on blocking keys.
There is no exact key match that can be used to extract candidate records from
the database. The load process needs to create a dataset containing unique
blocking keys, instead of unique SSKs. This set of unique blocking keys is
loaded into the aforementioned temporary table.
SCD Type 2
For SCD Type 2, you still need to upload a set of unique SSKs. The record
version to be extracted is determined by the SQL statement that implements the
database join, as discussed in the next section.
Extracting the reference lookup dataset

Once the unique SSKs are uploaded to the temp table, the next job in the
sequence executes a database join between the temp table and the
corresponding target table (For example, a join between the Customers table
and the temp table containing unique Cust_SSK).
Figure 15-4 on page 275 shows an example where the target table contains 100
million records. The uploaded temp table contains less than a million records.
The result from the join is at most the same as the number of records in the
temporary table.
The number of records coming out of the join is certainly less than the number in
the temp table, because uploaded SSKs are not present in the target table yet.

100 Million <=1 Million Could include
Target TMP Rows additional Where
Rows
Table SSKs predicates, such as
on CurrentFlag or
EffectiveDate for
DB Join
SCD Type 2
DSLink19
Join_Temp_CustSSKs Existing_CustSSKs
Figure 15-4 Extracting existing customer SKs with a database join
The join in Figure 15-4 guarantees that no useless data is extracted for the given
run. Only what is necessary (that is, referenced by any of the input files) is
extracted.
QS matching
As described in the previous section, for QS matching the temp table would have
been loaded with a set of unique blocking keys.
For QS Matching, the database join extracts the set of candidate records for all
input records, which serves as reference input for a QS Matching stage.
SCD Type 2
For SCD Type 2, the database join implements a filter to restrict records based
on one of the following elements:
򐂰 Current flag
򐂰 Expiration Date
򐂰 Effective Date

When to use full table extracts
For extreme situations when it is unavoidable to deal with large refresh datasets,
you might have to resort to full table extracts instead of the techniques described
so far.
That is the case in the following circumstances:

򐂰 The amount of input data is too large when compared with the size of the
target table.
For instance, the target table contains 100 million records, and the input file
has 90 million.
򐂰 The database join does not resort to an index-based plan, it does full table
scans, and the overhead of uploading unique keys into a temp table + plus
running a join inside the database is greater than extracting the whole table.
This makes it even more important to resort to Joins. The assumption is we are
dealing with large amounts of data, so performing a full table scan is much more
efficient.
15.6.5 Correlating data

Once the input and reference datasets are prepared (that is, the reference
dataset is restricted to a minimally relevant subset), one of the following options
can be used to do the actual record correlation:
򐂰 ChangeCapture
򐂰 Compare
򐂰 Difference
򐂰 Join plus Transformer
– Field-by-field comparisons in Transformer expressions
– Checksum comparison
򐂰 Normal Lookups
򐂰 SCD
Chapter 10, “Combining data” on page 149, reviews stages such as

ChangeCapture, Difference Compare, Checksum and SCD. They are described
in detail in the Parallel Job Developers Guide.
The SCD stage embeds logic specific to the processing of slowly changing
dimensions. When using the SCD stage, the data for the reference link must still
be restricted only to what is relevant for a given run. The SCD stage receives in
its reference input whatever the corresponding source stage produces.

The preferred, scalable options are as follows:
򐂰 Joins
The Join brings rows together and a Transformer implements the comparison
logic.
򐂰 Comparison stages
ChangeCapture, Difference, Compare
In Figure 15-5 on page 278 we show an example of using a Join to bring input
and reference records together based on the SSK.
The join type is left outer join. If there are incoming SSKs that were not found in
the target database table (that is, they are not present in the
Existing_Cust_SSKs dataset), the join still writes those records out. The
subsequent Transformer directs matching and non-matching records to separate
branches.
Non-matched records are directed to a surrogate key generator. All records, with
new and existing SKs are saved to the same dataset (Resolved_Cust_SSKs).
The top part of the example figure shows using a join to bring input and reference
records together based on the SSK.
The dataset containing the result from the database join (Existing_Cust_SSKs)
might have to include not only the existing SK values, but other fields as well.
This would be the situation, for example, where business requirements dictate
that other data values be compared to determine whether or not a given record
should be updated.
The Transformer after the join could compare checksum values. When doing
comparisons on checksums, the following tasks must be performed:
򐂰 A checksum value must be stored in the database table.
򐂰 The input dataset must be augmented with a newly created checksum value
on the same set of columns used originally to create the checksums stored in
the target table.

Figure 15-5 Resolving customers with a Sorted Join

The following options must be used only when the reference data always fits
comfortably in memory:
򐂰 Normal Lookups
򐂰 SCD stage
Figure 15-6 depicts an example of a job using a normal lookup.
Figure 15-6 Resolving customers with a normal Lookup
Jobs similar to the ones described would have to be implemented to resolve

Product SKs.
In Figure 15-7 on page 280, we describe how to use the resolved surrogate key
datasets for Customers and Products to resolve foreign key references in the fact
table (Sales).
For simplicity reasons, we are assuming that sales records are never updated.
Records are always added to the target table. The update process follows the
same approach as outlined for the Customer table.
The reference datasets are already sorted and partitioned. The input Sales file
needs to be re-partitioned twice: first in the input to the first sort (for Customer
lookup) and then in the input to the second sort (for Product lookup).

Figure 15-7 Resolving sales foreign keys

Sales records that do not find a corresponding Customer or Product SK are
directed to a reject dataset. This reject dataset might be accumulated to the input
to the next processing cycle (by then, hopefully, the missing customers and
products have arrived).
15.6.6 Keeping information server as the transformation hub

Assume an example where input data must be first loaded into an system of
record (SOR) database. This is a normalized database.
After the SOR is loaded with the day’s incremental load, the DW database must
be updated.
One might think of adopting a path as depicted in Figure 15-8. That diagram
represents a scenario in which the second database is updated with information
extracted from the first database.
Figure 15-8 Avoiding cascading database loads

That is what must be avoided, but unfortunately is something that has been
implemented in the real world.
Instead, Information Server must be kept as the transformation hub. Any data
that is fed into the second and subsequent databases is derived from datasets
stored locally by Information Server as persistent datasets or filesets. This is
 Minimal amount of queries

during batch processing
 Lookup/existence checks
SOR
Flat Information
Files Server
 N days of input files DW
 Loadable daily results

Persistent  Intra-day intermediate results
Datasets  Shadow of subset of SOR and DM contents
 Tables
 Aggregations
Figure 15-9 Information Server as the transformation hub
There is minimal interaction with the database. The SOR database is only
queried to the extent of obtaining restricted reference datasets as outlined in the
previous section.
Information Server should keep as datasets, for instance, rolling aggregations for
a limited number of days, weeks and months. Those datasets act as shadows to
content in the second and subsequent DBs in the chain.
15.6.7 Accumulating reference data in local datasets

When there is a high chance that resolved SKs are referenced again in the same
batch window, the reference datasets containing those resolved SKs must be
accumulated locally for the same processing window. This is the case when the
reception of inter-related input files is spread throughout the processing window.
One good example is banking. Most likely there is separate feeds during the
same processing window that relate to each other. For instance, new customers
are received at one time, and then later the account movements are received as
well. In this case, why first load and then re-extract the same data? Save the new
customer SKs in a local dataset or file set, and use this one later on for
subsequent files.

In Figure 15-10 on page 284 we show the collection of unique SSKs from
multiple files. It can be expanded to include a join against a locally stored dataset
containing the resolved Customer SKs for the previous run. This is characterized
by the segment highlighted inside the rectangle.
This approach avoids uploading to the temp table SSKs that were already
resolved by the previous run.

Figure 15-10 Isolating incoming SSKs, suppressing already resolved ones

Figure 15-11 shows how the existing SKs dataset is funneled with the resolved
dataset from the previous run (CURRENT_Resolved_Cust_SKs). The job
produces a new dataset (NEW_Cust_SKs) which, after the entire batch
sequence finishes, is appended to CURRENT_Resolved_SKs along with
Existing_SKs.
This way, we avoid reloading and re-extracting SSKs that were already resolved
by the previous run. We combine newly created SKs and newly extracted SKs
into a single dataset that serve as input to the next run.
Figure 15-11 Resolving customer SKs, taking into account already resolved ones
15.6.8 Minimize number of sequence runs per processing window

Jobs and sequences can be executed only once per processing cycle. If that is
not possible (that is, the transmission of input files is scattered throughout the
execution window) the DS sequence should wait for as many input files to be
received as possible before relaunching the entire sequence.
The goal is to minimize the impact of the startup time for all jobs. You must strike
a balance between the urgency to process files as they arrive, and the need to
minimize the aggregate startup time for all parallel jobs.

15.6.9 Separating database interfacing and transformation jobs
Provisioning and extraction stages must be contained in separate jobs. This
reduces the length of time database connections are kept open, and prevents
database loads or extract queries from failing in the event transformations fail for
other reasons. All loads to target tables must be delayed towards the end of the
processing window.
15.6.10 Extracting data efficiently

In this section we discuss techniques to consider to speed up the extraction of
data in bulk mode.
Use parallel queries

Whenever possible, use parallel queries. The availability of this mode depends
on the type of database stage and the partitioning of the source database table.
The ODBC stage can be set to run a query in parallel, depending on the number
of partitions on which the stage is run. The stage modifies the SQL query with an
extra where clause predicate based on a partition column.
The Oracle Enterprise stage requires one to set the partition table property.
The DB2 UDB Enterprise stage requires DB2 DPF to be installed on the target
database, to take advantage of direct connections to separate DB2 partitions.
DB2 connectors do not need DPF installed on the target database, but the
degree of parallelism of the query is still determined by the partitioning scheme of
the source table.
See Chapter 13, “Database stage guidelines” on page 189 for details on how to
do parallel reads with the various database stage types supported by DataStage.

Making use of database sort order
It is a good practice to make use of the sort order of the result set from a
database query. This is illustrated in Figure 15-12. The developer must work
closely with the database administrator to understand whether the query returns
a result set that conforms to the sort order needed in the job flow. In this
example, the query result is fed into one of the inputs to a Join stage.
Figure 15-12 Making use of the sort order of a database query result
If the output from the query is naturally sorted according to the criteria necessary
to perform the join, one can set properties as marked in red in Figure 15-12.
By naturally sorted we mean the query itself does not include an ORDER BY
clause, but still returns a properly sorted result set. This might happen, for
instance, if the database query plan involves a scan on index pages and the
index definition matches the sort order required in the DS job flow.
However, if the result set is not naturally sorted, a sort must be performed in one
of two places:
򐂰 In the DS parallel job
򐂰 Inside the database by means of an ORDER BY clause

Using DataStage is the natural choice.
However, depending on the nature of the source database (whether it is a

parallel database with enough spare hardware resources), you can think of
adding an ORDER BY clause to the query. Whenever doing so, the technique
depicted in Figure 15-12 on page 287 is used.
15.6.11 Uploading data efficiently

Parallel database stages invariably support the two major write methods:
򐂰 Bulk load
򐂰 Upsert
The first method uses highly optimized APIs to load large volumes of data the
fastest possible way. The underlying mechanisms tend to be specific to each
database, such as Oracle, DB2, and Teradata. With bulk loads, no SQL
statements are specified and they tend to be much faster than Upserts.
Bulk loads have the option of turning off or temporarily disabling indices and
constraints. These can be rebuilt and re-enabled at the end of the load process.
The Upsert method implies the execution of SQL statements, either generated
automatically or specified by the user as a custom SQL. Upserts rely on
database call-level interfaces and follow a record-at-a-time processing model, as
opposed to bulk loads. Most database CLIs support the execution of SQL
statements by sending arrays of rows to the DB.
Upserts invariably require indices to be present and enabled, otherwise, the cost
of executing an update or delete statement with a where clause is too inefficient,
requiring a full table scan per row.

Figure 15-13 shows the Write Mode options of a DB2 Connector stage:
򐂰 Update
򐂰 Delete
򐂰 Insert then Update
򐂰 Update then Insert
򐂰 Delete then Insert
򐂰 Bulk Load
Figure 15-13 DB2 Connector write model
Figure 15-14 presents the table action options for the DB2 Connector, when the
Write Mode is set to Bulk Load.
Figure 15-14 DB2 Connector table actions
Although DataStage presents a single stage type with different options, these
options map to underlying Orchestrate operators and database load utilities at
runtime.

Separate stages types have small variants, but drop-down options tend to be
mostly similar. See Chapter 13, “Database stage guidelines” on page 189 for
details on uploading with the database stage types supported by DataStage.
The question we address in this section is how to decide which database load
method to use for a given task at hand? The answer depends on a number of
factors:
򐂰 Size of the input DS versus size of the target table
򐂰 How clean the data to be uploaded is and if there might be any rejects
򐂰 Ratio of inserts X updates
To help determine the database load method to use, we provide the following key
selection criteria:
򐂰 Bulk Loads
– All rows of the input dataset are new to the target DB
– There is an option of disabling indices and constraints:
• Used when the cost of re-enabling indices and constraints is less then
the cost of updating indices and evaluating constraints for each and
every input row.
• Used when the data is guaranteed to be thoroughly cleansed by the
transformation phase, so when indices and constraints are re-enabled,
they do not fail.
򐂰 Upserts
– Required whenever the following is needed:
• An existing record must be removed (Delete mode);
• An existing record must be replaced (Delete then Insert mode);
• An existing record must be updated.
– For new records:
The following options are available:
• Insert
• Update then Insert
• Insert then Update
Must be used in the event the data is not sufficiently clean (for instance,
there is a chance new records might already be present in the target
table). Duplicate and violating records can be caught through reject links;
You must consider doing Bulk Loads for new records instead

– For updates to existing records:
The following options are available:
• Insert then Update mode
#inserts > #updates (the number of records to be inserted is larger then
the number of records to be updated);
• Update then Insert mode
#updates > #inserts (the number of records to be updated is larger
than the number of records to be inserted).
In the event the data is not sufficiently clean (that is, there is a chance the
new records might already be present in the target table), invalid records
can be caught through reject links.
The use of Upsert mode for updates is the natural way with DataStage database
stages.
Bulk updates
There is one additional alternative that can be explored for applying bulk updates
to target tables, assuming the incoming update records are guaranteed to be
clean. Perform the following steps:
1. Bulk load the update records to a database temporary (or scratch) table
a. This table should have neither indices nor constraints;
b. The assumption is the input update records are cleansed and are
guaranteed not to throw any SQL exceptions when processed against the
target table. This cleansing must have been accomplished in the
transformation phase of the ETL process;
2. Execute a single Update SQL statement that updates the target table from the
records contained in the temporary table.
This approach is illustrated in Figure 15-15.
(2) Update from Target

Temp table Table
InputUpdateRecords
Temp
(1) Bulk
UpdateRecords BulkLoad_Temp
Load
Table
Figure 15-15 Combination of bulk load and SQL statement for processing bulk updates

The advantage of this approach is that it would bypass the inherent overhead of
database call-level interfaces when submitting updates (which includes setting
up and sending to the database server incoming rows in the form of arrays, and
checking the return status for each and every array element).
The Bulk Update technique avoids that overhead because it assumes the data
does not raise any SQL exceptions. There is no need for checking the return
status on a row-by-row basis. This technique requires a closer cooperation with
DBAs, who have to set up the necessary table spaces for the creation of the
temp/scratch tables. Hopefully, the target database can do the two steps in
parallel.

16
Chapter 16. Real-time data flow design

Much of the DataStage history has focused on batch-oriented applications. Many
mechanisms and techniques have been devised to process huge amounts of
data in the shortest amount of time, in the most efficient way, in that environment.
We are aware that there are concerns related to limiting the number of job
executions, processing as much data as possible in a single job run, and
optimizing the interaction with the source and target databases (by using bulk
SQL techniques). All these are related to bulk data processing.
Batch applications tend to be bound to time constraints, such as batch execution

windows, which are typically require at least a few hours for daily processing.
However, there are several applications that do not involve the processing of
huge amounts of data at once, but rather deal with several small, individual
requests. We are not saying the number and size of requests are minimal. They
might actually be relatively large, but not as much as in high volume batch
applications.
For high volume scenarios, users typically resort to batch-oriented techniques.

So, what is discussed in this chapter must not be used to implement the
processing of daily incremental DW loads, for instance.
The types of applications in this chapter are focused on scenarios where there is
large number of requests, each of varying size, spanning a long period of time.

The following list details these types of applications:
򐂰 Real-time transactions
– B2C applications
There is a person waiting for a response, such as in a web-based or online
application.
– B2B applications
The caller is another application.
򐂰 Real-time reporting/decision making
– There are few or no updates to database tables.
– Involves fetching and massaging large amounts of data.
– Generates support of a specific decision, such as insurance approval.
򐂰 Real-time data quality
– Other applications invoke QualityStage for cleansing/de-duplication.
– An example is the use of MDM.
򐂰 Integration with third party business process control systems
– There is a business process coordinator that orchestrates the processing
of a series of steps, involving multiple participants. One of the participants
might be DataStage, which receives requests and posts results back to
the orchestrator through JMS, as an example. A third party example is
Tibco.
򐂰 Near-Real-time
– Message-oriented for instant access and update.
– Data trickled from a source mainframe to populate active data warehouses
and reporting systems
There are two capabilities that have been added to DataStage over time, to
address the needs of real-time:
򐂰 MQ/DTS (Distributed Transaction stage)
MQ/DTS addresses the need for guaranteed delivery of source messages to
target databases, with the once-and-only-once semantics.
This type of delivery mechanism was originally made available in DataStage
7.5, in the form of the UOW (Unit-of-Work) stage. The original target in DS 7.5
was Oracle. In InfoSphere DataStage 8.X, this solution has been substantially
upgraded (incorporating the new database Connector technology for various
database flavors) and re-branded as the Distributed Transaction stage.

򐂰 ISD (Information Services Director)
ISD enables DataStage to expose DS jobs as services for service-oriented
applications (SOA). ISD supports the following types of bindings:
– SOAP
– EJB
– JMS
DTS and ISD work in different ways and serve different purposes. In this section
we discuss the specifics of each. However there are common aspects that relate
to both:
򐂰 Job topologies
Real-time jobs of any sort need to obey certain rules so they can operate in a
request/response fashion.
򐂰 Transactional support
– Most real-time applications require updating a target database. Batch
applications can tolerate failures by restarting jobs, as long as the results
at the end of processing windows are consistent.
– Real-time applications cannot afford the restarting of jobs. For each and
every request, the net result in the target database must be consistent.
򐂰 End-of-wave
– The parallel framework implements virtual datasets (memory buffers) that
are excellent for batch applications. They are a key optimization
mechanism for high volume processing.
– Real-time applications cannot afford to have records sitting in buffers
waiting to be flushed.
– The framework was adapted to support End-of-Waves, which force the
flushing of memory buffers so responses are generated or transactions
committed.
򐂰 Payload processing
– Frequently batch applications have to deal with large payloads, such as
big XML documents.
– Common payload formats are COBOL and XML.
򐂰 Pipeline parallelism challenges
Fundamental for performance, the concept of pipeline parallelism introduces
challenges in real-time, but those can be circumvented.
We do not go into the details of the deployment or installation of DTS and ISD.
For that, the reader must see the product documentation.
Chapter 16. Real-time data flow design 295

16.1 Definition of real-time
The notion of real-time is not related to that which is typically used for embedded
systems, such as avionics, traffic control, and industrial control. DS is not for
embedded applications or for any type of scenario in which there are stringent
timing constraints.
The term real-time applies to the following types of scenarios:

򐂰 Request/Response
– As an example, there is a caller waiting for a response. They expect to
receive a response quickly, but there is no physical tragedy, such as a
transportation accident, that would occur if the response takes too long. It
instead might be that the caller receives a timeout.
– Typical examples:
• B2C, such as Web applications
• Online data entry applications
• B2B
– The following DS solutions can be applied:
• ISD with SOAP or EJB binding.
• ISD with JMS Binding, when the caller expects the response to be
placed on a response queue.
The term near-real-time has been used, and applies to the following types of
scenarios:
򐂰 Message delivery
– The data is delivered and expected to be processed immediately.
– Users can accept a short lag time (ranging from seconds to a few
minutes).
– There is no person waiting for a response.
– Examples:
• Reporting systems
• Active warehouses
– The following DS solutions can be applied:
• MQ->DTS/UOW
• ISD with Text over JMS binding
The high cost of starting up and shutting down jobs demanded that DataStage be
enhanced with additional capabilities to support these types of scenarios. This is
because their implementation with batch applications is not feasible.

16.2 Mini-batch approach
A common scenario in the field is the attempt to use batch applications for near
real-time scenarios (as defined in the previous section). In other words, instead
of having perpetual jobs waiting for requests or messages to arrive, systems
integrators and developers propose the use of batch jobs.
The idea is to accumulate input messages or requests to a certain extent, and

fire a job sequence. The hope is that such job sequences can finish in a short
amount of time. A common job duration of this type is 15 minutes, for instance.
These are the main problems with this approach:

򐂰 Extremely short execution windows
Application can be complex and as a result jobs tend to be huge. The final
result is that the total amount of time to start up and shut down all jobs in the
mini-batch sequences does not fit in the planned time window.
򐂰 Target database Inconsistencies
– The transactional consistency of the target database might be
compromised. Batch jobs might resort to multiple database stages spread
across multiple jobs. If the DS application is the only one updating the
target database, impact might be minimized. However, if other
applications update the same DB, there is a chance mini-batch updates
conflict with other real-time applications.
– Users see inconsistent views of the target database as the various
mini-batch provisioning jobs apply updates to target tables. These updates
are not part of a single unit-of-work and therefore the database state is not
as consistent (even though database constraints such as parent-child
relationships are maintained). For instance, a user might see an order
header without its details. This can only be guaranteed by an effective
unit-of-work.
In summary, we strongly advise against the use of mini-batches. Instead, use

either DTS or ISD, depending on the nature of the interaction with the sources
and targets.
16.3 Parallel framework in real-time applications

The advantages of the parallel framework are alive in the real-time world.
Service-oriented applications (SOAs), represented in their vast majority by

application server-based implementations such as J2EE, consist of a number of

services that are executed in a single thread of execution. Application servers
can spawn lots of threads on behalf of incoming requests, but each request is
serviced by a set of instructions that are executed one at a time in a strict order.
This execution model fits well for cases when there are absolutely no
interdependencies between the requests, and the methods that implement the
services are simple. As the picture becomes more complicated, this model
reaches its limits, and that is where the parallel framework has an advantage.
Pipeline and partitioning parallelism allow for the breaking up of the application
logic into several concurrent steps, and execution paths. This is depicted in
Figure 16-1.
Figure 16-1 Pipeline and partitioning parallelism
There are at least two scenarios in which the parallel framework helps break
through the limitations of a single-threaded execution model:
򐂰 Services of increased complexity
In real-time decision-support applications (such in the sales of insurance
policies), the amount of input and output data is small, but the service itself
must fetch, correlate, and apply rules on relatively large amounts of data from
several tables across data sources. This might involve doing lots of database
queries, sorts/aggregations, and so forth. All this processing takes a
considerable amount of time to be processed by a single threaded enterprise
JavaBean method. With the parallel framework, the several steps cannot only

be spread across multiple steps in a pipeline, but also have segments of the
logic executing across multiple partitions.
򐂰 Interdependencies and strict ordering in message-oriented applications
There are cases when a strict order must be imposed for the processing of
incoming messages, for example when bank account information is
transferred over to a decision support database. Account transactions must
be applied in the exact order in which they are executed in the source system.
In this scenario, the maximum throughput is determined by the efficiency of a
single thread of execution. In these types of scenarios, partitioning parallelism
is not applicable. However, these types of scenarios can greatly benefit from
breaking up the logic into multiple pipelined steps. Instead of having a single
thread of execution, there are multiple threads, each one executing on a
separate OS process.
Interesting challenges are introduced by pipeline parallelism in the real-time

world. However, the advantages exceed the challenges by far.
16.4 DataStage extensions for real-time applications

The parallel framework was extended for real-time applications. Three extended
aspects are:
򐂰 Real-time stage types that keep jobs always up and running,
򐂰 End-of-wave
򐂰 Support for transactions in target database stages.
16.4.1 Always-on source stage types

The ability for a job to stay always-on is determined by the nature of the source
stages. In batch applications, source stages read data from sources such as flat
files, database queries, and FTP feeds. All those stages read their corresponding
sources until exhaustion. Upon reaching the end-of-file (EOF), each source
stage propagates downstream the EOF to the next stage in the pipeline. In the
Figure 16-1 on page 298, the source dataset passes on the EOF marker to the
Transformer stage, which in turn passes it on to the sort, until it reaches the final
output dataset. The job terminates when all target stages run to completion.
In real-time applications, this model is not efficient, as discussed in 16.2,

“Mini-batch approach” on page 297. The job must remain running even though,
there might not be any data for it to process. Under those circumstances, the
stages are to remain idle, waiting for data to arrive.

There are a few stages that keep listening for more incoming data, and only
terminate under certain special circumstances, depending on the type of stage:
򐂰 ISD Input
This is the input stage for Information Services Director applications. It is the
same source stage, for a number of supported bindings (such as EJBs,
SOAP, JMS)
򐂰 MQConnector
– MQ reader stage in Information Server 8.X
– For pre-8.X Information Server, the corresponding stage was named
MQRead
򐂰 Custom plug-ins
New stage types can be created for special purposes, using the following
approaches:
– Custom operators
– Buildops
– Java client
The termination of real-time DS jobs is started by each source stage:

򐂰 For ISD jobs, the Information Services Director application must be
suspended or un-deployed.
򐂰 For applications, the MQConnector must reach one of the following
conditions:
– It reads a message of a certain special type (configured on the stage
properties).
– It reaches a read timeout.
– It reads a certain pre-determined maximum number of messages.
16.4.2 End-of-wave
The ability for jobs to remain always-on introduces the need to flush records
across process boundaries.
The parallel framework was originally optimized for batch workloads, and as
such, implements the concept of buffers in the form of virtual datasets. These
buffers are an important mechanism even for real-time jobs. However, if for a
given request (either an MQ message or an ISD SOAP request) there are not
enough records to fill virtual dataset buffers, they can wait until more requests
arrive so the buffers are flushed downstream.

An End-of-wave (EOW) marker is a sort of hidden record type that forces the
flushing of records. An EOW marker is generated by the source real-time stage
depending on its nature and certain conditions set as stage properties.
Figure 16-2 shows an example of how EOW markers are propagated through a
parallel job. The source stage can be an MQConnector for the purpose of this
illustration (in this case, the target stage is a DTS/Distributed Transaction stage).
EOW markers are propagated as normal records. However, they do not carry
any data. They cause the state of stages to be reset, as well as records to be
flushed out of virtual datasets, so a response can be forced.
Figure 16-2 EOWs flowing through a parallel job
The ISD and MQConnector stages generate EOW markers the following way:
򐂰 ISD Input
– It issues an EOW for each and every incoming request (SOAP, EJB, JMS).

– ISD converts an incoming request to one or more records. A request might
consist of an array of records, or a single record (such as one big XML
payload). The ISD input stage passes on downstream all records for a
single incoming request. After passing on the last record for a request, it
sends out an EOW marker.
– The mapping from incoming requests to records is determined when the
operation is defined by means of the Information Services Console.
򐂰 MQConnector
– It issues an EOW for one or more incoming messages.
This is determined by a parameter named Record Count.
– It issues an EOW after a certain amount of time has elapsed.
This is determined by a parameter named Time Interval.
– A combination of the two.
EOWs modify the behavior of regular stages. Upon receiving EOW markers, the
stage’s internal state must be reset, so a new execution context begins. For most
stages, (Parallel Transformers, Modify, Lookups) this does not have any practical
impact from a job design perspective.
For database sparse lookups, record flow for the Lookup stage is the same. But
the stage needs to keep its connections to the database across waves, instead
of re-connecting after each and every wave. The performance is poor.
However, there are a few stages whose results are directly affected by EOWs:
򐂰 Sorts
򐂰 Aggregations
For these two stages, the corresponding logic is restricted to the set of records
belonging to a certain wave. Instead of consuming all records during the entire
execution of the job, the stage produces a partial result, just for the records that
belong to a certain wave. This means that a sort stage, for instance, writes out
sorted results that are sorted only in the wave, and not across waves. The stage
continues with the records for the next wave, until a new EOW marker arrives.
16.4.3 Transaction support

Support for real-time applications is not complete without proper handling of
database transactions. Batch applications rely on standard database stage types
to execute bulk loads or inserts/updates/deletes.
The transaction context for database stages prior to Information Server 8.5
always involved a single target table. Those stages support a single input link,

which maps to one or a couple of SQL statements against a single target table. If
there are multiple database stages on a job, each stage works on its own
connection and transaction context. One stage is totally oblivious to what is going
on in other database stages.
This is depicted in Figure 16-3.
Figure 16-3 Transaction contexts with standard database stage types
The grouping of records into transactions might be specified as a stage property.

However, from a database consistency perspective, ultimately the entire batch
process tends to be the transactional unit. If one or more jobs fail, exceptions
must be addressed and jobs must be restarted and completed so the final
database state at the end of the batch window is fully consistent.
In real-time scenarios, restarting jobs is not an option. The updates from a

real-time job must yield a consistent database after the processing of every
wave.
Pre-8.1, Information Services Director (ISD) and 7.X RTI jobs had no option
other than resorting to multiple separate database stages when applying
changes to target databases as part of a single job flow. The best they could do
was to synchronize the reject output of the database stages before sending the
final response to the ISD Output stage (See 16.7.4, “Synchronizing database
stages with ISD output ” on page 353 for a discussion on this technique).
The Connector stages, introduced in Information Server 8.0.1 and further

enhanced in 8.5, provide for a new uniform interface that enables enhanced

transaction support when dealing with a single target database stage type, as
well as when dealing with message-oriented and heterogeneous targets (DTS).
The Connector and DTS stages are discussed in the following subsections.
Connector stages
Information Server 8.5 has Connector stages for the following targets:
򐂰 DB2
򐂰 Oracle
򐂰 Teradata
򐂰 ODBC
򐂰 MQSeries
IS 8.5 Connectors support multiple input links, instead of a single input link in
pre-8.5 connectors and Enterprise database stage types. With multiple input
links, a Connector stage executes all SQL statements for all rows from all input
links as part of single unit of work. This is depicted in Figure 16-4 on page 305.
One transaction is committed per wave. It is up to a source stage (such as ISD

input or MQConnector) to generate EOW markers that are propagated down to
the target database Connector.

Figure 16-4 Multi-statement transactional unit with a connector for a single database type
With database Connectors, ISD jobs no longer have to cope with potential
database inconsistencies in the event of failure. ISD requests might still have to
be re-executed (either SOAP, EJB, or JMS, depending on the binding type), but
the transactional consistency across multiple tables in the target database is
guaranteed as a unit of work.
For transactions spanning across multiple database types and those that include
guaranteed delivery of MQ messages, one must use the Distributed Transaction
stage, which is described next.
Distributed Transaction stage (DTS)

The stage that provides transactional consistency across multiple SQL
statements for heterogeneous resource manager types in message-oriented
applications is the DTS.

This stage was originally introduced in DataStage 7.5, with the name of
UnitOfWork. It was re-engineered for Information Server 8, and is now called
DTS. It is depicted in Figure 16-5.
Figure 16-5 Distributed Transaction stage
The DTS is discussed in detail in 16.6, “MQConnector/DTS” on page 313. It

applies to target databases all the SQL statements for incoming records for all
input links for a given wave, as a single unit-of-work. Each wave is independent
from the other.
It was originally intended to work in conjunction with the MQConnector for

message-oriented processing. When used in conjunction with the MQConnector,
it provides the following capabilities:
򐂰 Guaranteed delivery from a source queue to a target database
No messages are discarded without making sure the corresponding
transactions are successfully committed.
򐂰 Once-and-only-once semantics
No transaction is committed more than once.

In IS 8.5, the DTS supports transactions against single or heterogeneous
resource managers as part of the same transaction.
򐂰 DB2
򐂰 Oracle
򐂰 Teradata
򐂰 ODBC
򐂰 MQ
DTS can also be used without a source MQConnector, such as in Information

Services Director jobs. One still needs to install and configure MQ, because it is
required as the underlying transaction manager.
16.5 Job topologies

In Chapter 15, “Batch data flow design” on page 259, we describe a series of
techniques to be adopted to optimize batch-oriented applications. Unfortunately,
one cannot have the exact same implementation applied to both real-time and
batch scenarios. A real-time application must obey certain constraints
(elaborated upon in this section).
Figure 16-6 on page 308 illustrates how a real-time job must be structured. The
examples in this chapter use WISD stages, but they might be replaced with the
MQConnector as the source and the DTS as the target. The same constraints
apply to both message-oriented and SOA jobs.

Figure 16-6 Valid real-time data flow with all paths originating from the same source
All the data paths must originate from the same source, the source real-time
stage (either WISD Input or MQConnector). Figure 16-6 uses concentric blue
dotted semicircles to illustrate the waves originating from the same source.
Real-time jobs might split the source into multiple branches, but they must
converge back to the same target: either a DTS or WISD Output stage.
There can be multiple waves flowing through the job at the same time. You
cannot easily determine the number of such waves. It is dependent on the
number of stages and their combinability. The larger the job, the higher the
number of possible waves flowing through the job simultaneously.
Although originating from the exact same source, EOW markers might flow at
various speeds through different branches. This is dependent on the nature of
the stages along the data paths. In the example of Figure 16-6, the upper branch
might be slower because it involves a remote Web Service invocation, which
tends to be significantly slower than the standardization on the lower branch.

However, the framework makes sure EOW markers are synchronized whenever
they reach join stages such as the Funnel. The term “Join” is being used loosely
and refers to any stages that accept multiple inputs and support a single output
stream link, such as Join, Merge, Funnel, ChangeCapture, and Compare.
The need to restrict flow designs so that all data paths originate from the same
source restrict the types of jobs and stages that can be made use of. There is
one exception to this rule that is discussed later (consisting of data paths that
lead to the reference link of Normal Lookups).
Figure 16-7 presents one example of an invalid flow. The Funnel stage has two
inputs, one of them originating from the WISD input stage. The second input
(marked in red waves) originates from an Oracle Enterprise stage. The
semantics of this layout remain undefined and as a result it cannot be adopted as
a valid real-time construct. All data paths entering the Funnel stage must
originate from the same source. This example also applies to other Join stages,
such as Join, Merge, and ChangeCapture.
Figure 16-7 Invalid real-time data flow with multiple Source stages

The query in the Oracle Enterprise stage of Figure 16-7 on page 309 is executed
only once for the duration of the entire job life. The Oracle EE stage never issues
an EOW marker, and no matter how many EOWs are issued by the WISD input
stage, the Oracle EE stage is totally oblivious to what is going on the blue
branch. The WISD EOWs in no way affect the behavior of the Oracle EE stage.
Figure 16-8 presents another scenario, similar to the one in Figure 16-7 on
page 309. The Merge stage has an input that originates from an Oracle
Enterprise stage. As stated before, Merge and Join stages are perfectly legal
stages in real-time job flows. However, this is the case in which records from two
separate sources are being correlated by a common key. In other words, this
Merge stage is acting as a lookup.
No!
Figure 16-8 Using a Sparse or Normal Lookup instead of a Join

To make the job of Figure 16-8 on page 310 real-time compliant, we need to
replace the Merge with a Normal Lookup stage. A Normal Lookup stage is a
two-phase operator, which means in its first phase, immediately after job start up,
the Lookup stage consumes all records from its reference link and builds a
lookup table that is kept in memory for the duration of the job. In the second
phase, the Lookup stage processes incoming records that arrive through its
stream input, performing the actual lookup against the in-memory table
according to the lookup criteria.
The flow in Figure 16-7 on page 309, although containing a separate branch that
does not originate from the same WISD input stage, is valid, because the output
from the Oracle Enterprise stage is read only once during the startup phase. For
the duration of the entire job, the lookup table remains unchanged in memory.
Therefore, one can have branches originating from independent sources, as long
as those branches lead to reference links of Normal Lookups. Such branches
can have multiple stages along the way and do not have to necessarily attach the
source stage directly to the reference link.
It is important to note that for Normal Lookups the in-memory reference table
does not change for the entire duration of the job. This means Normal Lookups
must only be used for static tables that fit comfortably in memory.
The same layout of Figure 16-8 on page 310 can be used for a Sparse Lookup, in
case the reference data changes frequently and the incoming records need to do
a Lookup against the always-on data in the target database.

For reference links of Normal Lookups, the opposite of the previously stated rule
applies. Figure 16-9 illustrates a case where the reference link receives records
from the a branch that originates from the source WISD stage.
Figure 16-9 Invalid use of a Lookup stage in a real-time job
The two-phase nature of the Normal Lookup stage renders this design invalid.
The lookup table is built only once, at stage startup. However, the flow requires
EOWs to flow throughout all branches. For the second EOW, the upper branch
remains stuck, as the Lookup stage does not consume any further records.
Therefore, this flow does not work.
16.5.1 Summary of stage usage guidelines

In summary, the following general stage usage guidelines apply to always-on,
real-time data flows:
򐂰 Normal Lookups
– Must only be used for lookups against static reference datasets.
– The reference link must always originate from an independent branch.
򐂰 Join stages
– Stages included in this list are Funnel, Join, Merge, Compare,
ChangeCapture.
– All input links for Join stages must originate from the source real-time
stage (WISD Input or MQConnector).

򐂰 Sparse Lookups
– Must be used whenever there is a need to lookup against non-static, live
reference data repositories;
– Sparse Lookups are not limited to returning a single record. They might
return multiple records, which is a valid construct for real-time jobs. The
returned records might be further processed by transformations,
aggregations, sorts, joins, and so forth.
– Sparse Lookups are one of the predominant stage types in real-time,
always-on jobs, along with Parallel Transformers.
16.5.2 ISD batch topologies

ISD jobs might also conform to two batch topologies, which are not the focus of
this chapter:
򐂰 A job with no ISD Input, and with one ISD Output;
򐂰 A job with no ISD Input and no ISD Output.
These ISD job types are re-started for each request and might contain any data
flow patterns, including the ones that are okay for batch scenarios. The only
difference between them is the first requires data flow branches to be
synchronized before reaching the ISD Output. However, they can both have
branches that originate independently of each other.
16.6 MQConnector/DTS
The solution for transactional processing of MQ messages was originally
implemented in DataStage 7.X as a combination of two stages:
򐂰 MQRead
򐂰 Unit-Of-Work
These stages provided the guaranteed delivery from source MQ queues to target
Oracle databases, using MQSeries as the transaction manager. Those stages
were re-engineered in Information Server 8, and are now known as the
MQConnector and DTS stages.
The main difference is that they now implement a plug-able solution, taking
advantage of DataStage Connectors and supporting multiple types of database
targets, such as DB2, Oracle, Teradata, ODBC and MQ.

This plug-ability is implemented behind the scenes and is not a design feature
available to users. Users cannot build new components by plugging in existing
connectors to build new components. Instead, it is a feature that eases the
engineering effort to extend the capabilities of database interfacing. This is
visible through the DTS stage that, starting in IS 8.5, allows for multiple target
stage types as part of a single transaction.
DataStage historically had a set of database stages, some of them from the
existing DS Server (referred to as Plug-in stages) as well as a set of native
parallel database operators (such as Oracle EE, DB2 EE, Teradata EE, and so
forth). IBM defined a new common database interface framework, known as
Connector stages, which is the foundation for a new set of database stages.
The following types of Connectors are available:

򐂰 Oracle
򐂰 DB2
򐂰 Teradata
򐂰 MQSeries
򐂰 ODBC
Information Server 8.1 supports a single database type as the target for DTS.
Information Server 8.5 supports different target types as part of the same unit of
work.
16.6.1 Aspects of DTS application development

There are three main aspects involved in the development and deployment of
DTS jobs:
򐂰 The installation and configuration of the MQConnector and DTS stages
This involves identifying any fixpacks or updates that need to be downloaded
and installed so the Connector stages become available, both on the client
and the server sides.
– Requires the installation of corresponding database clients and MQSeries
software.
– Requires configuring XA resource managers and building an xaswit file.
򐂰 The design of the MQ/DTS job flows
The rules for design topology in 16.6.4, “Design topology rules for DTS jobs”
on page 316 must be obeyed.

򐂰 The configuration of the runtime environment
This is based on the ordering and the relationships between incoming
messages. It determines the scalability of the implementation.
– Determining the underlying configuration of work queues and number of
concurrent partitions.
– Setting the maximum number of messages that can flow through a job
simultaneously.
The job design must focus on obtaining a job flow that addresses business
requirements and, at the same time, complies with the constraints laid out in
16.5, “Job topologies” on page 307 and 16.6.4, “Design topology rules for DTS
jobs” on page 316.
There might be cases when there are concurrency conflicts that prevent multiple
messages from flowing through the job at the same time. In other words, the job
must somehow hold back messages from the source queue until the message
that is currently being processed is committed to the database. The third bullet
point is further elaborated in 16.6.11, “Database contention ” on page 339 and
16.10, “Pipeline Parallelism challenges” on page 366. The job design tends to be
independent from the runtime configuration.
16.6.2 Reference documentation

In this document, we do not intend to fully reproduce and replace the information
contained in the following two documents:
򐂰 The DTS Functional Specification
This document describes the DTS functionality and a detailed description of
the work queue topologies.
򐂰 Setting Up and Running DTS Jobs
This document provides information about the following topics:
– Installing and configuring MQConnector and DTS.
– Configuring XA resource managers.
– Shows sample DTS jobs.

16.6.3 A sample basic DTS job
Figure 16-10 presents a sample DTS job, with a basic pattern that consists of
four parts:
򐂰 The source MQConnector
򐂰 The parsing of the message payload into a DS record format
򐂰 The transformation of the message content according to business rules
򐂰 The execution of the transaction against the target database with the DTS
stage
Real-time jobs tend to be quite large, but that is the basic pattern to which they
tend to adhere.
The Column Import stage is sometimes replaced with an XMLInput stage, but the
result of the second step is always a set of records with which the rest of the DS
stages in the job can work.
Figure 16-10 Sample DTS job
16.6.4 Design topology rules for DTS jobs

There must be one source MQConnector stage and one target DTS stage.
The following rules from 16.5, “Job topologies” on page 307 must be obeyed:
򐂰 All data paths must originate from the same source stage (the DTS
Connector).
The exceptions are reference links to Normal Lookups. These must originate
from independent branches.
򐂰 All data paths must lead to the same target DTS stage
There might be dangling branches leading to Sequential File stages, for
instance, but these are to be used only for development and debugging. They
must not be relied on for production jobs.

򐂰 Normal Lookups can only be used for static reference tables.
򐂰 For any lookups against dynamic database tables, Sparse database Lookups
must be used.
Sparse Lookups might return multiple records, depending on the business
requirements.
16.6.5 Transactional processing

The MQ/DTS solution provides the following benefits:
򐂰 Guaranteed delivery of messages
No messages are discarded unless the corresponding transactions are
successfully committed to the target database.
򐂰 Once-and-only-once semantics
– Transactions on behalf of incoming messages are executed only once.
– If a job fails in the middle of a transaction, the entire transaction is rolled
back, and the original message remains either in the source queue or a
so-called work queue.
– Once-and-only-once semantics are only available for XA-Compliant
resource managers:
• Oracle
• DB2
• MQ
򐂰 Non-XA-compliant resource managers
– There is guaranteed delivery, but transactions might have to be replayed
in the event of job restart after failure.
• Non-XA-Compliant RMs: Teradata and ODBC
• Job logic must be implemented so transactions are IDEMPOTENT
That is, if replayed once or more with the same input parameters, they
must yield the same result in the target DB.
What makes this possible is that the MQConnector and DTS stages interact with
each other by means of work queues.
Work queues are not visible on the DS canvas. They are referenced by means of
stage properties.

The diagram in Figure 16-11 shows the relation between the source and target
stages through the work queue.
DB2
src queue
DB2
Local MQ transaction
moves message from
source to work queue work queue Distributed XA transaction deletes
under syncpoint control message(s) comprising unit-of-work
from queue and outputs
corresponding rows to DB2 target
Figure 16-11 Transaction support with MQConnector and DTS stages
The MQConnector transfers messages from the source to a work queue through
a local MQ transaction. It also forwards the payload to the next stage through its
output link. Typically, the next stage is a Column Import or XMLInput stage, to
convert the payload to DS records.
Once forwarded downstream, the message is guaranteed to be safely stored in

the work queue. Depending on the setting in the source MQConnector, this stage
might issue an EOW for each message, or for a group of messages.
On the target side, the DTS receives records from transformation stages along
multiple data paths. There might have been Parallel Transformers, Sparse
Lookups, Joins, Merges, and so forth. Those rows are stored by DTS in internal
arrays.
Upon receiving an EOW marker (which was, again, originated from the source
MQConnector), the target DTS stage performs the following tasks:
1. Invokes the transaction manager (MQSeries) to begin a global transaction.
2. Executes the various SQL statements for the rows that arrived as part of that
wave according to a certain order.
3. Removes from the work queue one or more messages that were part of this
wave.
4. Requests the Transaction Manager to commit the transaction
This is a global transaction that follows the XA protocol, involving MQSeries
and the target DBs.

Figure 16-11 on page 318 illustrates the fact that there is a close cooperation
between the source and target stages to ultimately achieve the guaranteed
delivery and once-and-only-once semantics.
16.6.6 MQ/DTS and the Information Server Framework

It might not be clear at first how the Information Server layers and parallel jobs fit
together, so the following explanation is helpful.
Figure 16-12 illustrates how an MQ/DTS job fits in the Information Server
Framework.
WAS (Domain Layer)

DSEngine
DS IS Admin
Director Endpoint
ASB ASB
Adapter Agent OSH
Conductor
DB2,
Oracle,
Teradata,
ODBC
Parallel Job
(Section
Leaders,
Players…)
Request XA Target MQ
Path to activate a job MQ Queue Transaction Queue
Runtime Parent/Child Process Relationship Work MQ
Runtime Data Path Queue(s)
Figure 16-12 MQ/DTS and the Information Server framework

There are four major layers:
򐂰 The DS client
Represented by DS Director
򐂰 The domain (or services) layer
The WebSphere Application Server
򐂰 The engine
Composed by the DS Engine and the ASB Agent
򐂰 The parallel framework processes
These are the processes that are executed on behalf of a parallel job, doing
the actual data extractions, transformations and load.
The illustration does not include the XMeta layer, which is not relevant for this
discussion. The XMeta database is accessed directly by the domain layer, mostly
at design time.
There are three major paths depicted in the figure:

򐂰 Job activation
Denoted by dotted (blue) arrows.
򐂰 Parent/child relationship of runtime processes
Denoted by large dashes (green) arrows.
򐂰 Data paths
Denoted by solid (red) arrows.
There are several other processes and modules, so the figure and this
discussion could go down to a finely grained level of detail. We chose the
elements that are relevant for the level of abstraction that is appropriate for this
discussion.
Job activation
To activate a job, the operator user opens DS Director, selects a job or sequence
for execution, and issues the execute command. The DS Director client sends
the request to the domain layer. Inside the domain layer, an administration
endpoint forwards the request to the Application Service Backbone (ASB)
adapter, which is sent to the ASB agent and finally reaches the DSEngine (at this
moment, there is a dsapi_slave process, which is the DSEngine process that is
activated on behalf of a client connection). Note that we do not provide any
discussion on how DSEngine connections are established.

Parallel job activation
Upon receiving the execute command, the DSEngine spawns an OSH process.
This process acts as the conductor and parent of all parallel framework
processes that are launched on behalf of the job being started.
The dsapi_slave becomes the parent of the OSH conductor that, in turn, is the
parent of all other parallel processes, including the MQConnector and DTS
operators.
The parent/child relationship between the DSEngine and the OSH processes is
denoted by the dotted green line.
Job execution
The activated processes establish communication channels among the virtual
datasets, namely the virtual datasets.
The path followed by the actual data being processed is denoted by the small
dashes (red) line. The data flows entirely in the realm of parallel framework
processes. There is no participation by the domain layer (as opposed to
Information Services Director).
16.6.7 Sample job and basic properties

In this section we discuss basic properties for the MQConnector and DTS stages
that one would set to develop a working job. We use Figure 16-11 on page 318
as the example.
The reader should see the product documentation for a detailed description of all
stage and link properties for these two stages.
In our discussion, the pictures combine aspects of the job design, with the job
flow towards the top of the picture and stage and link properties at the bottom.
Associations between the job flow and stage/link properties are marked as blue
arrows. Comments on individual properties are marked in red. We believe this
provides for a more concise characterization of the properties that are otherwise
scattered throughout separate windows.

Linking the source and target stages
Figure 16-13 details properties that provide for the linking between the source
and target stages. Both stages need to be assigned the same values for the
following elements:
򐂰 Queue Manager Name
򐂰 Work Queue Name
Linking by means of a queue manager and a work queue is not apparent on the
design canvas. It is, instead, expressed as stage and link properties.
For the DTS connector, the Queue Manager Name can be set either on the stage
or link properties (it supports a single output link).
At runtime, the work queue name might be appended with a partition number,
depending on the chosen runtime topology. This is discussed in 16.6.8, “Runtime
Topologies for DTS jobs” on page 326.
Figure 16-13 Queue manager and work queue information in Source and Target stages

MQConnector output link properties
Figure 16-14 indicates the basic properties one would have to set on the output
link for the MQConnector:
򐂰 Source queue name
򐂰 Message wait options
򐂰 Transaction size
򐂰 Generation of EOW markers
The options in this example make the job wait indefinitely for messages to arrive
in a source queue SOURCEQ. Every message is written to the work queue as a
separate transaction. For each transaction, an EOW marker is sent downstream
through its output link.
Figure 16-14 Basic MQConnector output link properties

You can also set an end of data message type that causes the MQConnector to
terminate upon receiving a message of a certain type. The message type is a
value that is set as an MQ message header property. This message can be
either discarded or sent downstream and written downstream depending on the
“Process end of data message” option.
Figure 16-15 presents a minimal set of columns that must be present in the
output link for the MQConnector stage.
Figure 16-15 Output link metadata for the MQConnector stage
The message payload must be sufficiently large to contain the largest expected
message. The APT_DEFAULT_TRANSPORT_BLOCK_SIZE environment
variable might have to be adjusted accordingly.
Each and every output row must have a field containing the MQ message ID
(DTS_msgID). For the MQConnector, this field can assume any name.
MQConnector knows it must assign the message ID to this field by means of the
data element property, which must be set to WSMQ.MSGID.

However, the DTS stage requires the message ID to be present in each of its
input links with the name DTS_msgID, so it is better to name the field this way
from the start.
There are other MQ header field values that can be added to the record
definition, by selecting the appropriate type from the list of Data Element values.
Input link properties for the DTS stage

On the target side, Figure 16-16 highlights the minimal set of properties that you
would have to set for each and every input link to obtain a working DTS job:
򐂰 Database Connection
򐂰 Write Mode
򐂰 SQL Generation or User-Defined Statements
Figure 16-16 Basic DTS properties for input links

The Connector stages support strict checking of field type and length. The user
has the option to turn this checking off.
Figure 16-17 depicts an example of input link metadata for the DTS.
Figure 16-17 Input link metadata for the DTS stage
The rules for input link columns are as follows:

򐂰 There must be a column called DTS_msgID, which carries the MQ message
ID of the source message.
򐂰 Any column that does not correspond to a column in the target database must
have a name prefixed with DTS_. For example, DTS_MySortKey. This
typically applies to sort key columns which are not database columns.
򐂰 If your job makes use of job-generated failures, then one or more links should
have columns called DTS_IsRejected, defined as an Integer, and
DTS_Message, defined as a Char or Varchar.
16.6.8 Runtime Topologies for DTS jobs

The Runtime Topology refers to the following aspects:
򐂰 The number of possible partitions
򐂰 The interaction between the source MQConnector and target DTS stages
through one of two ways:
– Work queues
– Solely relying on the source queue, instead of using work queues

As stated in 16.6.1, “Aspects of DTS application development” on page 314,
these are runtime aspects that do not really affect the way jobs are designed.
The design of real-time jobs must focus on the business requirements, making
sure the job design follows the design rules outlined in 16.6.4, “Design topology
rules for DTS jobs” on page 316.
However, the scalability of the job represented in terms of whether it is possible

to run just one or multiple partitions is determined by the following characteristics
of the incoming messages:
򐂰 Ordering
Messages might have to be processed in the exact same order they were
placed in the source queue, and there is nothing you can do about it from a
design standpoint. This might be the case of banking feeds from a mainframe
operational database to a reporting system. The ordering of the messages
must be enforced.
򐂰 Relationships
There might be an indication as to whether messages are correlated.
Appendix A, “Runtime topologies for distributed transaction jobs” on page 375,

presents a detailed discussion on runtime topologies. In this section we discuss a
few of them that make use of work queues. There is a possibility of bypassing
work queues, in the sense that the DTS removes messages directly from the
source queues. There are challenges with that. Use of work queues avoid those
potential issues.
For each topology, we discuss the corresponding stage and link properties.
No ordering, no relationships
Appendix A, “Runtime topologies for distributed transaction jobs” on page 375
presents the ideal case, from a scalability standpoint, which is illustrated in
Figure 16-18 on page 328. For as long as there are no ordering constraints and
no relationships between the incoming messages, the job can scale across
multiple partitions. There are multiple MQConnectors and multiple DTS
instances.
All MQConnectors read messages off the same source queue. However, they
transfer messages to work queues that are dedicated to each partition.
Upon job restart after failure, the job must be restarted with the exact same
number of nodes in the config file. Messages are either reprocessed out of the
private work queues, or read from the source queue.
All source MQConnectors transfer messages into individual work queues by

means of MQ local transactions.

Each target DTS stages applies a separate global transaction, involving the
target resource managers and the partition-specific private work queue.
3,1
MQ DT
Conn. WQ.0 Stage
3,1
SQ
4,3,2,1
4,2
MQ DT
Conn. WQ.1 Stage
4,2
Figure 16-18 A scalable topology, with no ordering and no relationships between

messages
Figure 16-18 indicates the properties that must be set in a parallel job to enable
the topology of Figure 16-19 on page 329:
򐂰 APT_CONFIG_FILE must point to a config file containing mode than one
node in the default node pool.
򐂰 The MQConnector and DTS stages must point to the same work queue name
(WORKQ).
򐂰 The target DTS stage must have the “Append node number” option set to Yes.
The MQConnector stage, knowing it is running on multiple partitions, appends

the partition number to the work queue name. There is no additional setting on
this stage to make this happen.

The DTS, on the other hand, does not append the partition number automatically
when it runs in parallel. That is because it is also intended to support the
topology described in “No strict ordering with inter-message relationships” on
page 332. For the DTS to interact with multiple work queues (that is, one per
partition) the user must set the property “Append node number” to Yes.
Figure 16-19 MQ/DTS job properties for fully parallel topology

Strict ordering
Unfortunately, the more common scenario is the one that requires the strict
ordering of messages, possibly without indication of how to group incoming
messages based on any sort of criteria.
This is the typical case when transferring data from operational mainframe-based
banking applications to reporting databases. Database transactions must be
executed in a strict order. Otherwise, the result in the target reporting database
does not make any sense when compared to the original source.
When that is the case, adopt the topology of Figure 16-20. There is a single node
in the configuration file and as a consequence, there is a single work queue.
Figure 16-20 Strict ordering for message processing
Figure 16-21 on page 331 presents the properties to enable the topology of
Figure 16-20:
򐂰 APT_CONFIG_FILE must point to a config file containing a single node in the
default node pool;
(WORKQ)
򐂰 The target DTS stage must have the option Append node number set to No.
Even though the job execution is restricted to a single partition, such applications
can still benefit from pipeline parallelism. That is because the processing is
broken down into separate processes. That is one of the major advantages of the
parallel framework.
The job designers use the same DS constructs to express logic that meets
business rules regardless of partitioning. At runtime, the framework still makes
seamless use of multiple processes, scaling much better than an equivalent
single-threaded implementation, which is typically the case with J2EE based
applications.

J2EE application servers are multi-threaded. However, unless a great deal of
effort is put into producing a sort of parallelism, individual Java methods that
implement SOA services in J2EE applications tend to be single-threaded.
Figure 16-21 MQ/DTS job properties for strict ordering

No strict ordering with inter-message relationships
There is one intermediate topology, for cases when there is no strict global
ordering, but there is a relationship between messages. Messages can be
grouped by a certain criteria specified in the form of a hash key.
Messages with the same hash key are still processed in the same order in which
they are posted on the source queue.
Figure 16-22 depicts a topology of no ordering, with relationships.
Figure 16-22 No ordering, with relationships

Figure 16-23 shows properties that must be set to enable the topology of
Figure 16-22 on page 332:
򐂰 APT_CONFIG_FILE must point to a config file containing a multiple nodes in
the default node pool;
(WORKQ);
򐂰 The target DTS stage must have the option Append node number set to No;
򐂰 The source MQConnector stage must execute in sequential mode;
򐂰 The input link for the stage immediately after the MQConnector (the column
import) must set hash partitioning.
Figure 16-23 MQ/DTS job properties for inter-message relationships

Bypassing work queues
There are other topologies discussed in Appendix A, “Runtime topologies for
distributed transaction jobs” on page 375 that bypass work queues. There are
potential issues, so we suggest using work queues.
16.6.9 Processing order of input links

The default behavior of the DTS stage is to process the rows in the order the
input links are defined in the DTS stage. Figure 16-24 presents the order of input
links to the DTS stage, which can be accessed from the Link Ordering tab.
The SQL statements for all rows sent to the first input link are processed, and the
SQL statements for all rows to the second input link are processed, and so on.
The order of processing of the records from input links are highlighted in the box
in Figure 16-24, which is pointed to with an arrow.
All SQL statements are processed as part of the same transaction. Input links for
parent tables must be listed first, followed by the links for children, or dependent
tables, so the referential integrity in the target database is maintained.
Figure 16-24 Default input processing order
There might be cases when this processing order is not adequate. The designer
might choose to adopt a different order: instead of processing all records on a

link-at-a-time basis, the DTS stage processes records ordered across all input
links.
Figure 16-25 shows an example in which there is a field named DTS_SeqNum.

This field must be present on all input links. Remember, because this field is not
present in the target tables, it must contain the prefix DTS_ according to the rules
described in “Input link properties for the DTS stage” on page 325.
The DTS stage properties that cause records to be processed in a certain order
are highlighted in the box shown in Figure 16-25. The order is maintained across
all input links. The DTS stage then processes records from different input links,
according to the order of the DTS_SeqNum values.
Figure 16-25 Cross-link ordering
16.6.10 Rejecting messages

There are two types of failure that can occur during the processing of work items:
򐂰 A work item is rejected if a single record in any of the rows comprising that
unit of work are flagged with the DTS_IsRejected column set to True. In this
event, there might be a job-provided error message in the DTS_Message

column. Any stage in the job (typically a transformation stage), might set the
DTS_IsRejected column to True if it detects any condition that identifies the
work item as warranting rejection.
򐂰 A failure occurs in outputting a record to a target. An example of this is an
error being returned by the database due to a constraint violation.
Figure 16-26 shows how you can enable the use of a work queue in the DTS
stage. The DTS stage properties that send messages for failed units of work to a
reject queue REJECTQ are highlighted by the box in Figure 16-26. You have the
option to stop the job upon the first error.
Figure 16-26 Enabling a reject queue in the DTS stage
In the event of a failure or rejection of a unit of work, the unit is rolled back so that
no data is written to the targets. Additionally:
򐂰 The source messages might be as follows:
– Moved to a reject queue (the default behavior)
– Left on the source queue
򐂰 The job might be aborted after a specified number of units of work have been
rejected.

򐂰 An option is provided in the DT stage to preface the rejected message data
with information about the failure. This information is provided in a 256 byte
record defined in Table 16-1.
Table 16-1 Failure information

Offset Length Type Description
int32 The link number containing the failure. If there are multiple, only the link
number for the first occurrence is listed.
int32 Status. Has one of the following values:

0 – Success: this particular message succeeded, but other messages in
this transaction failed.
1 – Unknown: the records in this message were not executed because of
an earlier failure (see StopOnFirst option)
2 – Target error: an error occurred writing to a target
3 – Commit error: the transaction cannot be committed
4 – Rejected: the transaction was rejected by logic in the job (the
DTS_IsRejected flag was set to true)
int32 Count of the number of rows in error or rejected in this message.
12 12 string The error code from the resource (if the status of the unit of work was 2)
24 236 string The error code from the resource (if the Status is 2), or the text message
from the DTS_Message column (if the Status is 4). In the latter case, this
contains the text from the first DTS_Message column encountered for this
message that is not null or an empty string.
270 242 string The text message from the DTS_Message column. This contains the text
from the first DTS_Message column encountered for this message that is
not null or an empty string.
If there are multiple failures, then the status reflects the status of the first
detected error. The count field provides a count for the number of rows in the
message that failed.

A property of the DT stage determines whether the processing of a unit-of-work
stops upon the first detection of an error. This results in the status of subsequent
messages in the unit-of-work being set to Unknown. If it is preferable to
determine whether these subsequent messages are unknown or not, the stage
can be configured to process all rows in the unit-of-work to obtain accurate
message status. Doing this has a negative impact on performance, because all
rows must be processed and subsequently rolled back.
Job-induced reject
The job RejectedByJobLogic demonstrates one way to reject a transaction by
detection of an error condition, such as an error in the data. The job looks like
Figure 16-27.
Figure 16-27 Reject by detection
There are three links to the DTS. Two of the links are to insert or update DB2
tables, and the third is purely for sending reject signals to the stage.

Though the DTS_isRejected column can be specified on any input link, it is often
convenient to have a dedicated reject link containing this column. This is the
approach used by this example. It is necessary for the link to execute a benign
statement that has no impact on the target. In this example, this is accomplished
by using a dummy table and a statement that has no effect by specifying the
where clause as WHERE 0=1. That table is depicted in Figure 16-28.
Figure 16-28 Reject by detection dummy table
16.6.11 Database contention

Batch jobs tend to eliminate database contention by using various techniques to
minimize the interaction with target and source databases. Batch applications
must use bulk load and unload techniques, decoupling the transformation
phases from extraction and provisioning jobs.
Real-time jobs cannot afford the luxury of minimizing the interface with target
databases. Lookups, for instance, tend to be Sparse (the exception being
lookups against static tables). Transactions must be successfully committed for
every wave.
DTS jobs are tightly coupled with databases. There are multiple queries and
update/insert/delete statements being executed concurrently, for separate
waves, and for records that are part of the same wave.
As a result of the multitude of SQL statements being executed for multiple

records in multiple waves across multiple partitions, there can be multiple
database exceptions, such as lock timeouts and deadlocks.

Though the parallel framework enables scalability and parallelism, it introduces
challenges that must be addressed for real-time database interoperability:
򐂰 Key collisions
򐂰 Deadlock and lock timeouts
These two challenges pertain to both ISD and MQ/DTS jobs. Key collisions are
discussed in 16.10, “Pipeline Parallelism challenges” on page 366.
Although lock-related exceptions can occur in both ISD and MQ/DTS jobs, they
are more pronounced in the latter.
Figure 16-29 presents an example of a DTS job that upserts records into three
separate tables. One might think that one upsert does not affect the others and
the job shall work perfectly well.
Figure 16-29 A simplistic view of database interactions
This might be true in a development and test environment, with only a few
messages submitted to the source queue at a time. Things go pretty well, up until
several more messages are submitted to that same source queue.

Upon submitting several messages, the developer might see deadlock and lock
timeout exceptions in the DS Director log, thrown by both the DTS and Sparse
Lookups. A more detailed view of what might happen at the database level can
be seen in Figure 16-30. Inserting records into dependent/children tables
involves not only placing locks on the table itself, but also checking foreign keys
against the parent table primary key index.
Figure 16-30 Database contention
This type of problem tends to be more pronounced when testing real-time jobs
against empty or small tables. In this case, most of the activity is going against
the same table and index pages, increasing the chance of contention. As there
are more records being fed into the pipeline by the source MQConnector, these
exceptions increase in number.
Having lock-related exceptions during real-time jobs is something we must to

avoid at all costs.
These exceptions are further aggravated when running the job with multiple
partitions. They even occur with a Strict Ordering topology (See “Strict ordering”
on page 330).

The solution is to limit the depth of the work queue. Set the Monitor Queue Depth
property to Yes and the Maximum Depth property to 1, as in Figure 16-31. In
addition, start with a single node in the config file.
Figure 16-31 Limiting the work queue depth
Setting the work queue depth to 1 and running the job on a single partition might
be done as a starting point. The queue depth and the number of partitions can
then be increased, as long as the lock-related exceptions do not occur.
There are number of aspects related to the database that require close attention
by the DBAs:
򐂰 Make sure the SQL plans are fully optimized and take advantage of indices.
򐂰 Avoid sequential scans in SQL plans.
򐂰 Make sure the locking is set at the record-level, not at page-level.
򐂰 Enable periodic execution of update statistics.
򐂰 DBAs must closely monitor database statistics and query plans.
In summary, real-time jobs tend to be closely coupled with databases. Therefore,

there must be a close cooperation with DBAs, to make sure the database is fully
optimized.

16.6.12 Scalability
The scalability of real-time MQ/DTS jobs depends on a number of factors, some
of them previously discussed:
򐂰 Whether jobs can run in parallel or just sequentially
– Ordering
– Relationships between messages
򐂰 Length of wave
How many messages can be grouped into a single wave/unit-of-work.
򐂰 Size of arrays
An input link property for the DTS stage. Each link might have a different
setting.
The one that has most impact is the length of the waves. Having more than one
message in a single UOW is most beneficial because each 2-Phase Commit has
a significant overhead. A significant improvement can be seen as the length of
the wave is increased by adding more messages.
At some point, adding more messages will no longer provide a noticeable

performance increase. However, that is not possible in most circumstances.
UOWs must remain limited to a single incoming message in the majority of
cases. The only resort is to adjust the size of the arrays.
Also, one might use multiple partitions whenever possible, depending on

ordering constraints (See the discussion on runtime topologies in 16.6.8,
“Runtime Topologies for DTS jobs” on page 326 and Appendix A, “Runtime
topologies for distributed transaction jobs” on page 375).
16.6.13 Design patterns to avoid

In this section we discuss design patterns that, although valid, are to be avoided.
So-called mini-batches are to be avoided as well. They are not even considered
a valid real-time pattern, which is one of the reasons why they are not listed in
this section. The topic on mini-batches deserves a special topic on its own, and is
presented in 16.2, “Mini-batch approach” on page 297.
Using intermediate queues

Starting with IS 8.5, it is possible to implement MQ/DTS jobs with MQConnector
as a target. This would make it possible to write jobs that implement guaranteed
delivery from source to target MQ queues.

Adopt this type of technique whenever the source is in the form of MQ
messages, and the target is supposed to be consumed by an existing or
third-party application.
One example is the case of a retail bank that implemented a DS job to transfer
data from a source mainframe to a target system, using MQ Series as the
standard middleware interface. This is depicted in Figure 16-32. DataStage jobs
pick incoming messages, apply transformations, and deliver results to the target
queue.
Figure 16-32 MQ as Source and Target
We consider this an acceptable pattern. However, as real-time jobs tend to grow

quite large, developers might start thinking about using intermediate queues to
break up larger logic into smaller steps, using intermediate queues as a way of
safely landing data, guaranteeing against any loss of messages.
This is the scenario depicted in Figure 16-33 on page 345. The real-time DS
logic is broken into multiple smaller steps.
We might trace a parallel to this approach with the DS Server-based mentality of

breaking batch processing into multiple steps and frequently landing data to disk.
In the case of batch processes, breaking larger jobs into smaller ones leads to
higher disk usage, higher overhead, and longer execution times.
In the case of real-time processes, the net effect, in addition to higher overhead,
is a longer lag time until the DataStage results are delivered to the target queue
for consumption by the target system.

Figure 16-33 Using intermediate queues for real-time processing (not recommended)
Real-time jobs have a different profile of use of OS resources than batch jobs.
Batch jobs move and transform large amounts of data at once, incurring much
higher memory and disk usage.
Although real-time jobs normally have a larger number of stages (it is common to
see hundreds of them in a single MQ/DTS or ISD flow), they tend to deal with
smaller input data volumes. The overall memory usage for the data and heap
segment of parallel operators, as well as disk use, tends to be smaller.
The goal of real-time jobs from a performance standpoint must be delivering
results to the targets (result queues, target database tables, ISD outputs) as
quickly as possible with minimal lag time.
The approach depicted in Figure 16-33 works against that goal.

16.7 InfoSphere Information Services Director
The second way of creating live, real-time jobs is with the InfoSphere Information
Services Director (ISD). ISD automates the publication of jobs, maps, and
federated queries as services of the following types:
򐂰 SOAP
򐂰 EJB
򐂰 JMS
ISD is notable for the way it simplifies the exposure of DS jobs as SOA services,
letting users bypass the underlying complexities of creating J2EE services for the
various binding types.
ISD controls the invocation of those services, supporting request queuing and
load balancing across multiple service providers (a DataStage Engine is one of
the supported provider types).
A single DataStage job can be deployed as different service types, and can
retain a single dataflow design. DS jobs exposed as ISD services are referred to
as “ISD Jobs” throughout this section.
ISD is described in detail in the document IBM InfoSphere Information Services

Director, which is included in the standard Information Server documentation
installed on client workstations as part of the IS Client install. This document
focuses on design patterns and practices for development of live ISD Jobs.
Figure 16-34 on page 347 is reproduced from the ISD manual and depicts its
major components. The top half of the diagram shows components that execute
inside the WebSphere Application Server on which the Information Server
Domain layer runs. Information Server and ISD are types of J2EE applications
that can be executed on top of J2EE containers.
With version 8.5, Information Server is tightly coupled with WebSphere

Application Server, as is Information Services Director and its deployed
applications.
The bottom half of Figure 16-34 on page 347 presents components that belong
to the Engine Layer, which can reside either on the same host as the Domain, or
on separate hosts.
Each WISD Endpoint relates to one of the possible bindings: SOAP, JMS, or
EJB. Such endpoints are part of the J2EE applications that are seamlessly
installed on the Domain WebSphere Application Server when an ISD application
is successfully deployed by means of the Information Server Console.

Figure 16-34 Major components of Information Services Director
The endpoints forward incoming requests to the ASB adapter, which provides for
load balancing and interfacing with multiple services providers. Load balancing is
another important concept in SOA applications. In this context it means the
spreading of incoming requests across multiple DataStage engines.
A single DS engine can service a considerable number of requests. However, if

bottlenecks are identified, more engines might be added. In this case, the same
DS jobs must be deployed on all participating engines.
The ASB agent is a Java application that intermediates the communication

between the ASB adapter and the DS engine. This is a standalone Java
application, meaning it does not run inside a J2EE container. Because it is a Java
application, it is multi-threaded and supports the servicing of multiple incoming
requests.

An ASB agent implements queuing and load balancing of incoming requests.
Inside a given DS engine, there might be multiple pre-started, always-on job
instances of the same type, ready to service requests. That is where the
always-on nature comes into play. Instead of incurring in the cost of reactivating
jobs whenever a new request arrives, the ASB agent controls the life cycle of
jobs, keeping them up and running for as long as the ISD application is supposed
to be active.
One important concept to understand is the pipelining of requests. This means

multiple requests are forwarded to an ISD job instance, before responses are
produced by the job and sent back to the ASB agent. There is a correlation
between this concept and the pipeline parallelism of DS parallel jobs. There can
be multiple concurrent processes executing steps of a parallel job in tandem. The
pipelining of requests allows for multiple requests to flow through a job at the
same time.
For ISD applications, there is a direct mapping between a service request and a
wave. For each and every service request, an end-of-wave marker is generated
(See 16.4.2, “End-of-wave” on page 300).
In this section we include an explanation on how the components in Figure 16-34

on page 347 map to Information Server layers, and how they interact with each
other from various aspects:
򐂰 Job activation
򐂰 Parent/child process relationships
򐂰 The flow of actual data.
Figure 16-35 on page 349 illustrates these relationships.
Once an ISD job is compiled and ready, the ISD developer creates an operation
for that job using the Information Server Console. That operation is created as
part of a service, which is an element of an ISD application.
Once the ISD operations, services, and application are ready, the Information
Server Console can be used to deploy that application, which results in the
installation of a J2EE application on the WebSphere Application Server instance.
The deployment results in the activation of one or more job instances in the
corresponding service provider, namely the DS engines that participate in this
deployment.
This is represented by the dashed (blue) arrows. The ISD Administration

application forwards the request to the target DS engines through the ASB
adapter and the ASB agents.

Figure 16-35 Data and control paths in ISD applications
The DS engine, in turn, spawns one or more parallel job instances. A parallel job
is started by means of an OSH process (the conductor). This process performs
the parsing of the OSH script that the ISD job flow was compiled into and
launches the multiple section leaders and players that actually implement the
runtime version of the job. This is represented by the dashed (green) arrows.
Incoming requests follow the path depicted with the red arrows. They originate
from remote or local applications, such as SOAP or EJB clients, and even
messages posted onto JMS queues. There is one endpoint for each type of
binding for each operation.
All endpoints forward requests to the local ASB Adapter. The ASB adapter
forwards a request to one of the participating engines according to a load
balancing algorithm. The request reaches the remote ASB agent, which puts the
request in the pipeline for the specific job instance.

The ASB agent sends the request to the WISD Input stage for the job instance.
The ISD job instance processes the request (an EOW marker is generated by
the WISD Input for each request) and the job response is sent back to the ASB
agent through the WISD Output stage.
The response flows back to the caller, flowing through the same components
they came originally from. As opposed to the MQ/DTS solution, the WebSphere
Application Server is actively participating in the processing of requests.
16.7.1 The scope of this section

In this section we focus on design patterns and practices for development of
always-on ISD jobs.
The following aspects are outside of the scope of this document:

򐂰 A tutorial on SOA, Web Services, and J2EE development
򐂰 Federation, Classic Federation, and Databases (DB2, Oracle) as service
providers
򐂰 The steps related to the definition of applications, services, and operations
using the Information Server Console
򐂰 Batch jobs exposed as ISD services (that is, ISD jobs that do not have a
WISD Input stage).
For these aspects, the reader should see the Information Services Director
product manual.
For tutorials on SOA, Web Services and J2EE there are countless resources
available in books and on the web.
The reason we exclude ISD job topologies that are not of always-on type is
because those types of jobs should follow the recommendations outlined in
Chapter 15, “Batch data flow design” on page 259.

16.7.2 Design topology rules for always-on ISD jobs
These are the job flow design topology rules for always-on ISD jobs:
򐂰 There must be one source WISD Input stage and one target WISD Output
stage.
򐂰 The following rules from 16.5, “Job topologies” on page 307 must be obeyed:
– All data paths must originate from the same source stage (the WISD
Input). The exception is reference links to Normal Lookups. These must
originate from independent branches.
– All data paths must lead to the same target WISD Output stage.
– There might be dangling branches leading to Sequential File stages, for
instance, but these are to be used only for development and debugging.
They must not be relied on for production jobs.
– Normal Lookups can only be used for static reference tables.
– For any lookups against dynamic database tables, Sparse database
Lookups must be used.
– Sparse Lookups might return multiple records, depending on the business
requirements.

16.7.3 Scalability
There are two layers of scalability supported by ISD:
򐂰 Multiple job instances.
򐂰 Deploy across multiple service providers, that is, multiple DS Engines.
These are illustrated in Figure 16-36.
ISD Load
http App Balancing
server Server
DS Server
Job
instances
DS Server
ISD Load
Balancing
ASB Agent
Figure 16-36 ISD load balancing
Whether to deploy multiple job instances and multiple DS Engines to service

requests for the same ISD Application is a matter of performance monitoring and
tuning. The performance of the application must be closely monitored to
understand possible bottlenecks, which then drive the decisions in terms of
additional job instances and engines.
To address performance requirements, we recommend making assessments in

the following order:
1. Job design efficiency
2. Number of job instances
3. Number of DS Engines
The first task is to make sure the logic is efficient, which includes, among other
things, making sure database transactions, transformations and the entire job
flow are optimally designed.
Once the job design is tuned, assess the maximum number of requests that a
single job instance can handle. This is a function of the job complexity and the
number of requests (in other words, EOW markers) that can flow through the job

simultaneously in a pipeline. If that number of requests is not enough to service
the amount of incoming requests, more than one job instance can be
instantiated. In most cases, increasing the number of job instances helps meet
SLA requirements.
However, there might be cases when one reaches the maximum number of
requests a single ASB agent can handle. This means the limit of a single DS
engine has been reached. This can be verified when no matter how many job
instances are instantiated the engine cannot handle more simultaneous
requests. If this is the case, add more DS Engines either on the same host (if
there is enough spare capacity) or on separate hosts.
Keep in mind that throughout this tuning exercise, the assumption is that there
are enough hardware resources. Increasing the number of job instances and DS
Engines does not help if the CPUs, disks, and network are already saturated.
16.7.4 Synchronizing database stages with ISD output

Prior to Information Server 8.5, ISD jobs had to rely on multiple database stages
when updating more than one table in a single job flow, similar to what was
discussed in 16.4.3, “Transaction support” on page 302.
Figure 16-3 on page 303 illustrates database transaction contexts in a batch

application. Each separate database stage maintains a separate connection and
transaction context. One database stage is oblivious to what is going on in other
database stages even if they are part of the same job.
In an always-on job, the database stages must complete the SQL statements
before the response is returned to the caller (that is, before the result is sent to
the WISD Output stage).
Pre-8.5 jobs required a technique, illustrated in Figure 16-37, that involves using
a sequence of stages connected to the reject link of a standard database stage.
Figure 16-37 Synchronizing database and WISD output

The only exception to this rule in releases prior to 8.5 was the Teradata
Connector, which in version 8.0.1 already supported an output link along the
lines of what is described in 16.7.6, “ISD with connectors” on page 357. However,
prior to 8.5, the Teradata did not allow more than a single input link.
There is a Column Generator that created a new column, which was used as the
aggregation key in the subsequent aggregator. The output from the aggregator
and the result from the main job logic (depicted as a local container)
synchronized with a Join stage.
The Join stage guaranteed that the response is sent only after the database
statements for the wave complete (either successfully or with errors).
The logic implemented by means of the ColumnGenerator, Aggregator, and Join

stages were repeated for each and every standard database stage present in the
flow. Synchronized results made sure there was no other database activity going
on for a request that has already been answered.
Again, that is what had to be done in pre 8.5 releases. In IS 8.5, the database
Connectors are substantially enhanced to support multiple input links and output
links that can forward not only rejected rows, but also processed rows.
There are two alternatives for always-on 8.5 jobs when it comes to database
operations:
򐂰 DTS
򐂰 Database connector stages
These are discussed in the subsequent sections.
16.7.5 ISD with DTS

The DTS stage has been enhanced as part of the 8.5 release to support
transactions without the presence of source and work queues. This enables a
DTS stage to respond and commit transactions solely as a result of EOW
markers. Using the mode described in this section, a DTS stage can be included
as part of an always-on ISD job. The DTS stage supports units of work that meet
the following requirements:
򐂰 Multiple SQL statements
򐂰 Multiple target types
– DB2
– Oracle
– Teradata
– Oracle
– MQ Series

Figure 16-38 on page 356 illustrates a DTS stage used in an ISD job. It
significantly reduces the clutter associated with the synchronization of the
standard database stages as discussed in the previous section. This illustration
is provided to give an overall perspective.
The DTS stage supports an output link, whose table definition can be found in
category Table Definitions/Database/Distributed Transaction in the DS repository
tree.
When used in ISD jobs, the Use MQ Messaging DTS property must be set to
NO. Note that although source and work queues are not present, MQ Series
must still be installed and available locally, because it acts as the XA transaction
manager.
DTS must be used in ISD jobs only when there are multiple target database
types and multiple target database instances. If all SQL statements are to be
executed on the same target database instance, a database connector must be
used instead. This is discussed in the following section.

Figure 16-38 DTS in an Information Services Director job

16.7.6 ISD with connectors
The database connectors in IS 8.5 support units of work involving multiple SQL
statements against the same target database instance.
For connector stages, an output link carries the input link data plus an optional
error code and error message.
If configured to output successful rows to the reject link, each output record
represents one incoming row to the stage. Output links were already supported
by the Teradata Connector in IS 8.0.1, although that connector was still restricted
to a single input link.
Figure 16-39 on page 358 shows an example in which there are multiple input
links to a DB2 Connector (units of work with Connector stage in an Information
Services Director Job). All SQL statements for all input links are executed and
committed as part of a single transaction, for each and every wave. An EOW
marker triggers the commit.
All SQL statements must be executed against the same target database
instance. If more than one target database instance is involved (of the same or
different types), then the DTS stage must be used instead.
The example also depicts multiple input links to a DB2 Connector (units of work
with Connector stage in an Information Services Director Job).

Figure 16-39 Units of work with Connector stage

16.7.7 Re-partitioning in ISD jobs
Always-on ISD jobs can take advantage of multiple partitions, as long as all
partitions are collected back to a single partition prior to sending the result to the
ISD output stage.
Multiple partitions provide for a way of scaling a single job instance across
multiple partitions. This is a key advantage for complex services that receive
request payloads of an arbitrary size and, as part of the processing of those
requests, require the correlation of large amounts of data originating from sparse
lookups.
One such example is an application developed for an insurance company for the
approval of insurance policies. Jobs are of great complexity, with several
Lookups and lots of transformation logic. That type of job design benefits from
intra-job partitioning, instead of executing the entire job logic in a single partition.
With multiple partitions, there are multiple instances of stages along the way that
cut significantly the response time for such complex services.
As stated, all partitions must be collected back to a single partition, which

conveys the final response to the WISD Output stage.
As discussed in 16.7.3, “Scalability” on page 352, there can be multiple job

instance servicing requests, as determined at deployment time through the IS
Console.
As always, proper monitoring must be in place to make sure hardware resources

are not saturated and there is enough spare capacity to accommodate additional
partitions and job instances.
16.7.8 General considerations for using ISD jobs

In this section we provide some general considerations for when and how to use
ISD with always-on jobs.
򐂰 Generally speaking, the best job candidates for service publication with
Information Services Director are those with the following characteristics:
– Represent fine-grained data integration or data quality functions
– Are request/response oriented
– Have small to medium payloads
– Are entirely idempotent (can be run over and over again in succession
without impact)

– Do little or no writing
– Have high water mark traffic requirements that are 300 requests per
second or less
򐂰 For parallel jobs always use a single node config. The quantities of data used
in a service are typically small (industry guidelines), and do not need high
degrees of partitioning. Jobs are not easy to debug when deployed as
services. Having multiple nodes un-necessarily complicates things. The only
exception to this rule, which must be well qualified, is if the service takes little
input and returns little output, but generates large scale amounts of data for
intermediate processing. For instance, an internal set of lookups that
generate vast amounts of rows through Cartesian product would result in the
need for multiple nodes and parallelism in the midst of a service.
Helpful hint: Use node pools in the configuration file to restrict the default
node pool to one node, and have other node pools available so that stages
can run in parallel for the conditions described.
򐂰 Test your jobs fully without having them deployed as services. Use Flat File
Source to Flat File Target to perform QA on the logic and processing of the
job, or alternatively, RowGenerator (or in server, a leading Transformer with
Variable and Constraint stages). Replace the row generation stage and flat
files with WISDInput and WISDOutput once tested and verified.
򐂰 Consider using shared containers for critical logic that is shared among
classic style batch jobs and always on. Though not always possible because
of logic constraints, this offers the opportunity to keep the core logic in one
place for simpler maintenance, and only alter the sources and targets.
򐂰 Beware of binary data in character columns. Binary data, especially binary
00’s, is incompatible with SOAP processing and might cause problems for the
ISD Server, the eventual client, or both.
򐂰 Beware of DS Parallel settings that pad character fields with NULLs. This
usually only happens with fixed length columns. Varchar is a solution, as is
setting $APT_STRING_PADCHAR to a single blank (no quotes around the
blank).
򐂰 When performing INSERTs to relational databases, be safe and set array and
transaction sizes to 1 (one). Otherwise you might not see the new key values
immediately, or until the job is complete.
򐂰 Do not expect values written to flat file and other targets to be immediately
available or visible. Use a transaction size of 1 for such things.
򐂰 The text bindings (Text over JMS and Text over HTTP) exist primarily for
always on jobs, and because they have no formal metadata definition on their
payload, and work best for single-column input and output (XML is a common

content used for the buffers). These bindings are able to start jobs through the
other two topologies. However, their payloads (the message in the requesting
or input queue) cannot populate the job parameter input.
To populate individual columns and job parameters using one of the text
bindings you can use values in the JMS header columns (JMS headers,
custom headers, and HTTP header fields). Jobs with multi-column output can
also populate the payload and header information for a message in the output
or response queue (one column to the payload and the others populating
header properties).
򐂰 Job sequences cannot be published as services. Alternatively, publish a job
with WISD Input and WISD Output stages that contains a BASIC Transformer
(or use a server job) and invokes the utility sdk routine (found in the
Transformer category in the pull-down list of the Transformer) called
UtilityRunJob. This allows you to call another DS job from in your WISD Job.
It has the ability to wait, or immediately return after starting the remote job
sequence. Another approach is to create a server job with only (manually
written) job control logic. This is not considered a sequencer and can be
published as a service.
16.7.9 Selecting server or EE jobs for publication through ISD

When selecting server or EE jobs for publication through ISD, parallel jobs are
the preference because of the following reasons:
򐂰 Support for pipeline and partitioning parallelism.
򐂰 Must be the preferred engine for any new development.
򐂰 They can scale on cluster and grid environments, and are not restricted to run
on the head node only.
Reduce the strain on the head node and therefore it is less likely to require
the load balancing across multiple service providers (DSEngines);
򐂰 Transaction support enhancements described in 16.7.5, “ISD with DTS” on
page 354 and 16.7.6, “ISD with connectors” on page 357 overcome previous
limitations with parallel stages that made DS Server jobs necessary
The DTS and Database Connector stages in IS 8.5 provide for a similar
functionality to a DS Server combination of a Transformer stage directly
connected to an ODBC stage, for instance.
The following scenarios justify the publication of DS Server jobs as ISD services:
򐂰 Publishing existing DS Server jobs;
򐂰 Invocation of DS Sequences

– Sequences fall into the Batch category, and are therefore outside of the
scope of this chapter;
– Sequences cannot be exposed as ISD services;
– A DS Server job can be created with a Transformer that in turn invokes a
DS Sequence by means of the utility sdk routine (find it in the Transformer
category in the pull-down list of the Transformer) called UtilityRunJob.
16.8 Transactional support in message-oriented

applications
In this section we provide an overview of transactional support in message
oriented applications.
Information Server supports two types of solutions for message-oriented

applications:
򐂰 Information Services Director with JMS bindings;
򐂰 MQ/DTS.
The obvious difference is that the first is for interoperability with JMS, and the
second is for MQ.
One might use of ISD with JMS for the processing of MQ Series messages and
MQ/DTS for the processing of JMS messages by setting up a bridging between
MQ and JMS by means of WebSphere ESB capabilities. However, there is a
relatively high complexity in the setup of a bridging between JMS and MQ.
We put both solutions side-by-side in a diagram, Figure 16-40 on page 363. The
goal of this illustration is to draw a comparison of how transactions are handled
and the path the data flows.
Both ISD and MQ/DTS jobs are parallel jobs, composed of processes that
implement a pipeline, possibly with multiple partitions. The parent/child
relationships between OS processes are represented by the dotted green lines.
The path followed by the actual data is represented by solid (red) lines.
ISD jobs deployed with JMS bindings have the active participation of the
WebSphere Application Server and the ASB agent, whereas in an MQ/DTS job,
the data flow is restricted to the parallel framework processes.
MQ/DTS jobs provide both guaranteed delivery and once-and-only-once

semantics. This means a given transaction is not committed twice against the
target database, and each and every transaction is guaranteed to be delivered to

the target database. Transaction contexts are entirely managed by DS job
stages. There is a local transaction managed by the MQConnector (for the
transfer of messages from the source queue to the work queue) and an XA
transaction involving multiple database targets and the work queue.
In an ISD job, there are transaction contexts managed by DS parallel processes,

depending on how many database stages are present in the flow. See 16.7.4,
“Synchronizing database stages with ISD output ” on page 353 through 16.7.6,
“ISD with connectors” on page 357.
Figure 16-40 Transactional Contexts in ISD and MQ/DTS jobs
However, there is one additional transaction context one the JMS side, managed
by EJBs in the WAS J2EE container as JTA transactions.
JTA transactions make sure no messages are lost. If any components along the
way (WAS, ASB Agent or the parallel job) fail during the processing of an
incoming message before a response is placed on the response queue, the

message is re-processed. The JTA transaction that picked the incoming
message from the source queue is rolled back and the message remains on the
source queue. This message is then re-processed upon job restart.
This means database transactions in ISD jobs exposed as JMS services must be
idempotent. For the same input data, they must yield the same result on the
target DB.
In summary, the strengths of the MQ/DTS and ISD/JMS solutions are as follows:
򐂰 MQ/DTS
Guaranteed delivery from a source queue to a target database.
򐂰 ISD/JMS
Adequate for request/response scenarios when JMS queues are the delivery
mechanism.
As discussed in 16.6, “MQConnector/DTS” on page 313, the DTS stage has

been enhanced in version 8.5 to support MQ queues as targets. Therefore, a
request/response type of scenario can be implemented with MQ/DTS. However,
this implies a non-trivial effort in setting up the necessary bridging between JMS
and MQ on the WAS instance.
There are a few important aspects that need attention when taking advantage of
ISD with JMS bindings:
򐂰 The retry attempt when JTA transactions are enabled is largely dependent
upon the JMS provider. In WebSphere 6, with its embedded JMS support, the
default is five attempts (this number is configurable). After five attempts the
message is considered a poison message and goes into the dead letter
queue.
򐂰 One problem is that there are subtle differences in all of this from provider to
provider. It becomes a question of exactly how things operate when MQ is the
provider, or when the embedded JMS in WAS is the provider.
򐂰 The JMS binding also creates a spectrum of other issues because of the pool
of EJBs. Multiple queue listeners can result in such confusions as messages
ending up out of order, and a flood of concurrent clients going into ISD, and
overwhelm the number of instances that you have established for DataStage.

16.9 Payload processing
There are a number of ways of processing incoming payloads:
򐂰 Regular records
– Possible with ISD.
– The ISD operation must be defined in such a way its signature is mapped
to records written out by the WISD Input stage.
– With regular records, no extra parsing is needed.
򐂰 XML
– The input MQ message, JMS message, or SOAP payload contains an
XML document as a varchar or binary field.
– The XMLInput stage must be used to parse the incoming message into
multiple records of different types in different output links.
– One might have to use a cascade of XMLInput stages depending on the
complexity of the input XML.
– XMLInput stages might have to be complemented with XSLT
transformations with XMLTransformer stages.
DS Parallel jobs can handle any type of payload, it is a matter of using the
right components for the tasks at hand.
򐂰 COBOL/Complex Records
– The input MQ message, JMS message, or SOAP payload contains a
complex record as a varchar or binary field.
– Either the Column Import or Complex Flat File stage must be used to
parse the payload.
– One might have to use a cascade of such stages, depending on the
complexity of the input record format.
If the input payload is too large, you might run into memory issues. The
environment variable APT_DEFAULT_TRANSPORT_BLOCK_SIZE must be
tuned accordingly.

16.10 Pipeline Parallelism challenges
As discussed in Chapter 15, “Batch data flow design” on page 259, DS batch
applications must be optimized to minimize the interaction with databases. Data
must be loaded and extracted in bulk mode, and the correlation of data should
avoid the use of sparse lookups.
As opposed to batch, real-time applications are tightly coupled with databases.

These types of jobs cannot use the same load/unload and correlation
techniques. The result is that they make heavy use of Sparse Lookups, so they
have immediate access to the latest information stored in the target DBs.
However, using Sparse Lookups is not enough to address real-time database
access needs.
The nature of the parallel framework introduces interesting challenges, that are
the focus of this section. This discussion applies to both MQ/DTS and ISD jobs.
The challenges result from pipeline parallelism, in which there are multiple
processes acting on a dataflow pipeline concurrently. The most notable
challenge is key collision, for which a solution is described in the next section.
16.10.1 Key collisions

Key collision refers to a condition that involves an attempt to create two or more
surrogate keys when processing multiple contiguous records that contain the
same business key.
Figure 16-41 illustrates two contiguous records (R1 and R2), containing the
same business key (“AA”) for which either an existing surrogate (SK) must be
returned from the database, or a new SK must be created if such SK does not
exist yet.
Target Table Next Val Target Table
Lookup SK Not Found Generate Insert

for Incoming New SK New SK Funnel
natural key R1 (AA, SK1)
R2 (AA)
Found
Figure 16-41 Back-to-back records leading to key collisions

The stages are represented by means of squares. They all execute as separate
processes.
R1 arrives and goes through the first lookup first. An SK is not found, so it is
forwarded to the Generate New SK stage. This one invokes a NextVal function in
the target database to obtain a new SK value that is from now on associated to
business key AA. The Generate New SK stage returns a new value, SK1, which
is now propagated downstream. This value, though, is not saved in the target
database, so for all purposes, AA’s surrogate is not SK1 in the database yet.
R2 arrives immediately next. It also goes through the same initial lookup, and
because AA is not saved yet in the target database, it is also forwarded to the
Generate New SK stage.
In Figure 16-42, you can see that both records were sent downstream to the
Insert New SK stage. R2 has been assigned a surrogate key value of SK2.
However, only the first one is inserted successfully. The second one violates a
unique key constraint and therefore from now on, R2 becomes invalid.
R1 (AA, SK1)
Target Table Next Val Target Table
R2 (AA, SK2)
Lookup SK Not Found Generate Insert

for incoming Funnel
New SK New SK
natural key
Found
Figure 16-42 Key collision occurs

There might also be situations in which there are inter-branch dependencies,
such as in Figure 16-43. One branch generates an SK value that is queried by a
different branch.
This inter-branch dependency can also be extrapolated to include different jobs

and even other types of applications, with all of them querying and updating
surrogate keys for the same target table. So, the question is how to solve this key
collision problem in a straightforward way.
Any solutions involving any type of caching or global lookup table in DataStage
would not solve this problem; it would only further complicate things. The Lookup
must always be done against the target database. The least a DataStage cache
would do is introduce the problem of keeping that cache in synch with the target
database, so this approach is not suitable.
Single DS Job
Logically Dependent Branches
Source Parse /
Input
DB2 Switch
MQ queue
Logs
Logical Dependency
Figure 16-43 Inter-branch dependencies
16.10.2 Data stubbing

The technique that allows for circumventing the key collision issue is called data
stubbing.
This technique involves creating a skeleton record in the target database which,
for a short period of time, holds the association between a business key and its
surrogate key. Upon full processing of a record through the application flow, the
record and its relationships to other records are fully populated.

The implementation of this technique comprises the following:
򐂰 A stored function
– This function implements the logic to look up an SK for an incoming
business key. If such an SK is not found, it creates a new SK and stores
that value in the target table, establishing the link between the business
key and the newly-created surrogate key.
– The logic is implemented as a single transaction.
– This function becomes the access point for all applications.
򐂰 A Sparse Lookup to invoke the stored function
All records that need to go through the lookup/SK generation process, invoke
the stored function by means of a sparse lookup.
This is illustrated in Figure 16-44. There is a DB2Connector stage attached to the

reference link of a Sparse Lookup stage. Every incoming record flows through
this stage and result in the invocation of this function, in this particular branch of
the job flow.
Figure 16-44 Invoking a stubbing UDF with a sparse lookup

The red square in Figure 16-44 on page 369 shows how a hypothetical function
_SUBFUNC_ is invoked. This example is for DB2 and it uses a special syntax as
illustrated in Figure 16-44 on page 369.
The UDF returns a column named SK, which is propagated downstream. This
column is defined in the output link for the DB2 Connector stage.
DB2 functions
The code for the sample DB2 UDF is listed in Figure 16-45. The function is
defined as ATOMIC.
It first queries the target table _TABLENAME_. If no value is found, it inserts a

new record, invoking the Nextval function on a database sequence for this table
in the insert’s value list. The final statement returns whatever value is present in
the table for the incoming business key at that point.
Figure 16-45 A DB2 UDF that implements a data stubbing function

The UDF is invoked by means of a select statement as illustrated in
Figure 16-46. It requires the use of the Cast function to pass in the arguments.
Figure 16-46 Sample query to invoke a stubbing DB2 UDF
Oracle stored functions

Figure 16-47 shows an equivalent Oracle stored function. The syntax is different,
but the logic is the same as the DB2 version.
Figure 16-47 An Oracle stored function that implements data stubbing
The Oracle stored function needs to be created with the PRAGMA

AUTONOMOUS_TRANSACTION option, so that it can be invoked in the
SELECT clause of a query. By default, select statements cannot invoke UDFs

that contain DML statements. That PRAGMA directive makes that possible. The
only catch is that the transaction inside the function is independent from the
transaction in which the calling SQL is being executed. That is acceptable,
because the sparse lookups run independently from the target DTS or database
Connectors anyway.
The syntax to invoke this Oracle stored function is shown in Figure 16-48.
Figure 16-48 Sample query to invoke a stubbing Oracle function
16.10.3 Parent/Child processing

Now that we have a notion of the parallelism challenges and solutions, we can
discuss the specific problem of handing parent/child relationships. The solution
builds upon the use of the previous technique.
The assumption is that the SKs are not generated automatically at the time of
insert. Instead, the SK can be obtained by invoking a stubbing UDF to obtain the
surrogate key for the parent table. This SK value is propagated downstream, in
both parent and children links. The DTS or target database Connector stage then
commits the records as a single unit of work.
16.11 Special custom plug-ins

A custom solution was implemented to support the integration of DataStage into
a Tibco application. The advantage of this solution is that it integrates DataStage
directly with a third-party JMS provider, without the need to set up bridges
between WAS JMS and the external JMS provider.
Also, the JMSPlug-in runs entirely outside of a J2EE container, making the
interoperability with the third-party JMS provider straightforward. This is
illustrated in Figure 16-49 on page 373.
This solution guarantees the delivery of a message to a target queue. It uses

techniques similar to the work queues in the MQ/DTS solution.
However, because the source and target JMSPlug-in stages run on standalone
JVMs, outside of J2EE containers, there is no support for JTA transactions.
Messages are never lost, but they might be re-processed in the event of restart

after failure. Therefore, database transactions along the job flow must be
idempotent.
This solution is described in the document Integrating JMS and DataStage Jobs
with the JMSPlug-in available from IBM Information Management Professional
Services.
Figure 16-49 The JMSPlug-in
16.12 Special considerations for QualityStage

QualityStage jobs developed for use under WISD need to be conscious of the
concept described in the previous section, especially when feeding data into a
reference match.

An often desired pattern is to feed the incoming service-based data as the
primary link for the reference match, and bring in reference data from a fixed data
source. This violates the rules described in 16.7.2, “Design topology rules for
always-on ISD jobs” on page 351.
Instead of having a fixed data source attached to the reference link, perform a
Lookup based upon your high level blocking factors in the incoming WISD
request.
This involves using a Copy stage that splits the incoming row, sending a row to
the primary input as before when sending the other row to a Lookup where
multiple reference candidates can be dynamically retrieved.
Be sure to make this a Sparse Lookup if you expect that the source data could
change when the WISD job is on (enabled), and check it carefully to be sure you
have set it to return multiple rows.

A
Appendix A. Runtime topologies for

distributed transaction jobs
There are several ways in which DT jobs can be deployed. The MQ Connector
and DT stage components need to offer flexibility so that the users can select the
appropriate topology for their particular use case. The choice of topology is
dependent upon the nature of the source data. Functional requirements that
influence the appropriate topology are as follows:
򐂰 Order
This requirement governs whether the messages have to be processed in the
order they are written to the source queue. If the order must be maintained,
then it is not possible to execute the job in parallel, because there is no
co-ordination between player processes on multiple nodes.
򐂰 Relationships
This requirement governs whether source messages are related, that is, they
have a key field that indicates they need to be processed as a unit. A hash
partitioner is used to ensure that all messages with a given key are processed
by the same node.
In the diagrams in this section the numbers under the queues represent
message sequence numbers and illustrate how messages might be distributed
across queues. Letters represent hash partitioning key fields. The solid arrows
show the movement of MQ messages to and from queues. The dashed lines

represent the job links. For clarity, only MQ and DT stages are shown in these
jobs, but in reality there are other stages in between to implement the business
logic of the extract, transform, and load (ETL) job.
A.1 No ordering, no relationships

If the order of processing of source messages is unimportant, and there is no
relationship between messages, then it is possible to run the jobs fully in parallel.
Each node contains an MQ stage, a work queue, and a DT stage. The MQ

stages access a single source queue and distribute these to the nodes. Because
the MQ stage instances are reading the messages destructively off the source
queue, there is no contention for messages. The work queues are required,
because the instances of the MQ Connector source stages must remove the
message from the source queue. These stages cannot browse the queue,
because all instances would then see and process the same messages. This is
depicted in Figure A-1.
Figure A-1 No order, no relationships

The reason to have multiple work queues is to restart jobs upon catastrophic
failure. Consider the case of a single work queue, depicted in Figure A-2.
Figure A-2 Unsupported topology
If a restart is required, each MQ Connector stage player process attempts to

browse messages from the single work queue. The MQ stages have to browse
the queue because they must leave the messages on the work queue. This
cannot be supported. It is not possible to have multiple processes browse the
same queue because they all read the same messages. With multiple work
queues, this problem does not arise, because each MQ process has exclusive
access to browse the queue in the event of a restart.
Multiple work queues also aid performance, because there is no contention for
work queues, and the DTStage is more likely to find its message from the head
of the queue.
Appendix A. Runtime topologies for distributed transaction jobs 377

A.2 No ordering, with relationships
If there is a need to ensure that all messages that are related to each other by a
shared key value are sent to the same node, a single MQ stage combined with
the use of a hash partitioner is used. This is depicted in Figure A-3.
Figure A-3 No ordering, with relationships
In this scenario, there is a single work queue, because the MQ Connector cannot
determine which node is targeted for a specific message. By contrast, the
MQRead operator is implemented as a combinable operator, and can be
combined with the hash partitioner. This permits MQRead to learn to which
partition the message is sent, and can therefore direct the message to the
appropriate work queue if multiple work queues are used. MQRead can therefore
use multiple work queues, but the MQ Connector cannot.
Whether there is a significant performance hit by using a single work queue

rather than multiple queues needs to be determined. Any impact is due to locking
mechanisms on the queue, and potentially having to search further in the queue
to match messages by ID.

Ordering
If there is a need to process the messages in the order they arrive on the source
queue it is necessary to execute the entire job sequentially. The reason for this is
that there is no synchronization between nodes, and so distributing messages to
multiple nodes cannot guarantee any ordering of messages. The topology would
appear as shown in Figure A-4.
Figure A-4 Strict ordering
A.3 Bypassing work queues

The scenarios depicted in section A.2, “No ordering, with relationships” on
page 378 and section “Ordering” on page 379 can be modified. It is possible to
configure both the MQ and DT stage to operate in absence of a work queue.
When running with no work queue, the MQ connector would browse the source
queue rather than destructively getting messages from the source queue. The
DT stage then accesses the same source queue and destructively get the
messages.
The revised topologies are depicted in Figure A-5 and Figure A-6 on page 380.
When no ordering is required, it is as depicted in Figure A-5.
Figure A-5 No work queues, no ordering
Appendix A. Runtime topologies for distributed transaction jobs 379

When ordering is required, it is as depicted in Figure A-6.
Figure A-6 No work queues, with ordering
One advantage of running without a work queue is that restart after failure is
simpler. The job can be restarted, and it continues from where it was aborted.
Additionally, evidence with the MQRead operator indicates that reading from a
source queue and writing to a work queue under sync point control for a small
transaction size (small number of messages) is an expensive operation. By
omitting the need to write to a work queue, the overall performance is improved.
There are dangers in this approach however. Prior work with MQ and
WebSphere TX determined two scenarios where source messages can be
missed due to the message cursor not detecting messages:
If multiple processes are writing to the source queue, the queue browser might
miss a message if the PUT and COMMIT calls from these processes are
interspersed in a certain order.
If the writing processes use message priorities, the queue browser does not see
messages of a higher priority, as they jump ahead of the current cursor position.
The solution offers support for all of these scenarios. It is the responsibility of the
job designer to select the appropriate settings to configure the stages to enable a
particular scenario.

B
Appendix B. Standard practices summary

In this appendix we summarize suggestions that have been outlined in this
document, along with cross-references for more detail.

B.1 Standards
It is important to establish and follow consistent standards in directory structures
for install and application support directories. An example directory naming
structure is given in 3.1, “Directory structures” on page 22.
Directory Structures
IBM InfoSphere DataStage requires file systems to be available for the following
elements:
򐂰 Software Install Directory
IBM InfoSphere DataStage executables, libraries, and pre-built components
򐂰 DataStage Project Directory
򐂰 Runtime information (compiled jobs, OSH scripts, generated BuildOps and
Transformers, logging info);
򐂰 Data Storage
– DataStage temporary storage: Scratch, temp, buffer
– DataStage parallel dataset segment files
– Staging and Archival storage for any source files
By default, these directories (except for file staging) are created during
installation as subdirectories under the base InfoSphere DataStage installation
directory.
In addition to the file systems listed, a DataStage project also requires a proper
amount of space in the Metadata layer (which is a relational database system).
As opposed to the Project directory, the Metadata layer stores the design time
information, including job and sequence designs, and table definitions.
This section does not include requirements and recommendations for other
Information Server layers (Metadata and Services) The discussion here is strictly
in terms of the Engine layer.
A single Information Server instance (services and metadata) can manage

multiple engines. The following discussion pertains to the setup of a single
engine instance.

Data, install, and project directory structure
This directory structure contains naming conventions, especially for DataStage
Project categories, stage names, and links. An example DataStage naming
structure is given in 3.2, “Naming conventions” on page 32.
All DataStage jobs must be documented with a short description field, as well as
with annotation fields. See 3.3, “Documentation and annotation” on page 47.
It is the DataStage developer’s responsibility to make personal backups of work

on local workstations, using the DS Designer DSX export capability. This can
also be used for integration with source code control systems. See 3.4, “Working
with source code control systems” on page 50.
B.2 Development guidelines

Modular development techniques must be used to maximize re-use of DataStage
jobs and components, as outlined in Chapter 5, “Development guidelines” on
page 69.
The following list is a summary of those development guidelines:

򐂰 Job parameterization allows a single job design to process similar logic
instead of creating multiple copies of the same job. The Multiple-Instance job
property allows multiple invocations of the same job to run simultaneously.
򐂰 A set of standard job parameters must be used in DataStage jobs for source
and target database parameters (DSN, user, password, and so forth) and
directories where files are stored. To ease re-use, these standard parameters
and settings must be made part of a designer job template.
򐂰 Create a standard directory structure outside of the DataStage project
directory for source and target files, intermediate work files, and so forth.
򐂰 Where possible, create re-usable components such as parallel shared
containers to encapsulate frequently-used logic.
DataStage Template jobs must be created with the following elements:

򐂰 Standard parameters (as examples, source and target file paths, and
database login properties).
򐂰 Environment variables and their default settings
򐂰 Annotation blocks
Job parameters must be used for file paths, file names, and database login
settings.
Appendix B. Standard practices summary 383

Parallel shared containers must be used to encapsulate frequently-used logic,
using RCP to maximize re-use.
Standardized error handling routines must be followed to capture errors and

rejects. Further details are provided in 5.6, “Error and reject record handling” on
page 74.
B.3 Component usage

As discussed in 5.7, “Component usage” on page 85, the following guidelines
must be used when constructing parallel jobs with DataStage:
򐂰 Never use Server Edition components (BASIC Transformer, Server Shared
Containers) in a parallel job. BASIC routines are appropriate only for job
control sequences.
򐂰 Always use parallel datasets for intermediate storage between jobs.
򐂰 Use the Copy stage as a placeholder for iterative design, and to facilitate
default type conversions.
򐂰 Use the parallel Transformer stage (not the BASIC Transformer) instead of
the Filter or Switch stages.
򐂰 Use BuildOp stages only when logic cannot be implemented in the parallel
Transformer.
B.4 DataStage data types

Be aware of the mapping between DataStage (SQL) data types and the internal
DS parallel data types, as outlined in Appendix H, “DataStage data types” on
page 423.
Use default type conversions using the Copy stage or across the Output
mapping tab of other stages.

B.5 Partitioning data
Given the numerous options for keyless and keyed partitioning, the following
objectives help to form a methodology for assigning partitioning:
򐂰 Objective 1: Choose a partitioning method that gives close to an equal
number of rows in each partition, and minimizes overhead. This ensures that
the processing workload is evenly balanced, minimizing overall run time.
򐂰 Objective 2: The partition method must match the business requirements and
stage functional requirements, assigning related records to the same
partition, if required.
Any stage that processes groups of related records (generally using one or
more key columns) must be partitioned using a keyed partition method.
This includes, but is not limited to the following stages: Aggregator, Change
Capture, Change Apply, Join, Merge, Remove Duplicates, and Sort. It might
also be necessary for Transformers and BuildOps that process groups of
related records.
In satisfying the requirements of this second objective, it might not be possible
to choose a partitioning method that gives close to an equal number of rows
in each partition.
򐂰 Objective 3: Unless partition distribution is highly skewed, minimize
repartitioning, especially in cluster or Grid configurations.
Repartitioning data in a cluster or grid configuration incurs the overhead of
network transport.
򐂰 Objective 4: Partition method must not be overly complex.
The simplest method that meets these objectives is generally the most
efficient and yield the best performance.
Using the above objectives as a guide, the following methodology can be applied:
򐂰 Start with Auto partitioning (the default).
򐂰 Specify Hash partitioning for stages that require groups of related records.
– Specify only the key columns that are necessary for correct grouping as
long as the number of unique values is sufficient.
– Use Modulus partitioning if the grouping is on a single integer key column.
– Use Range partitioning if the data is highly skewed and the key column
values and distribution do not change significantly over time (Range Map
can be reused).

򐂰 If grouping is not required, use Round-robin partitioning to redistribute data
equally across all partitions.
This is especially useful if the input dataset is highly skewed or sequential.
򐂰 Use Same partitioning to optimize end-to-end partitioning and to minimize
repartitioning:
– Being mindful that Same partitioning retains the degree of parallelism of
the upstream stage.
– In a flow, examine up-stream partitioning and sort order and attempt to
preserve for down-stream processing. This might require re-examining key
column usage in stages and re-ordering stages in a flow (if business
requirements permit).
Across jobs, persistent datasets can be used to retain the partitioning and sort
order. This is particularly useful if downstream jobs are run with the same degree
of parallelism (configuration file) and require the same partition and sort order.
Further details about partitioning methods can be found in Chapter 6,

“Partitioning and collecting” on page 91.
B.6 Collecting data

Given the options for collecting data into a sequential stream, the following
guidelines form a methodology for choosing the appropriate collector type:
򐂰 When output order does not matter, use Auto partitioning (the default).
򐂰 When the input dataset has been sorted in parallel, use Sort Merge collector
to produce a single, globally sorted stream of rows.
When the input dataset has been sorted in parallel and Range partitioned, the
Ordered collector might be more efficient.
򐂰 Use a Round-robin collector to reconstruct rows in input order for round-robin
partitioned input datasets, as long as the dataset has not been repartitioned
or reduced.
Further details on partitioning methods can be found in Chapter 6, “Partitioning

and collecting” on page 91.

B.7 Sorting
Using the rules and behavior outlined in Chapter 7, “Sorting” on page 115, apply
the following methodology when sorting in a DataStage parallel data flow:
򐂰 Start with a link sort.
򐂰 Specify only necessary key columns.
򐂰 Do not use Stable Sort unless needed.
򐂰 Use a stand-alone Sort stage instead of a Link sort for options that not
available on a Link sort:
– Sort Key Mode, Create Cluster Key Change Column, Create Key Change
Column, Output Statistics
– Always specify DataStage Sort Utility for standalone Sort stages
– Use the “Sort Key Mode=Do not Sort (Previously Sorted)” to resort a
sub-grouping of a previously-sorted input dataset
򐂰 Be aware of automatically-inserted sorts.
Set $APT_SORT_INSERTION_CHECK_ONLY to verify but not establish
required sort order
򐂰 Minimize the use of sorts in a job flow.
򐂰 To generate a single, sequential ordered result set use a parallel Sort and a
Sort Merge collector.
B.8 Stage-specific guidelines

As discussed in Section 9.1.1, “Transformer NULL handling and reject link” on
page 140, precautions must be taken when using expressions or derivations on
nullable columns in the parallel Transformer:
򐂰 Always convert nullable columns to in-band values before using them in an
expression or derivation.
򐂰 Always place a reject link on a parallel Transformer to capture / audit possible
rejects.
The Lookup stage is most appropriate when reference data is small enough to fit
into available memory. If the datasets are larger than available memory
resources, use the Join or Merge stage. See 10.1, “Lookup versus Join versus
Merge” on page 150.

Limit the use of database Sparse Lookups to scenarios where the number of
input rows is significantly smaller (for example 1:100 or more) than the number of
reference rows, or when exception processing.
Be particularly careful to observe the nullability properties for input links to any
form of Outer Join. Even if the source data is not nullable, the non-key columns
must be defined as nullable in the Join stage input to identify unmatched records.
See 10.2, “Capturing unmatched records from a Join” on page 150.
Use Hash method Aggregators only when the number of distinct key column
values is small. A Sort method Aggregator must be used when the number of
distinct key values is large or unknown.
B.9 Database stage guidelines

Where possible, use the Native Parallel Database stages for maximum
performance and scalability, as discussed in section 13.1.1, “Existing database
stage types” on page 190.
Native Parallel Database stages are as follows:

򐂰 DB2/UDB Enterprise
򐂰 Informix Enterprise
򐂰 ODBC Enterprise
򐂰 Oracle Enterprise
򐂰 Netezza Enterprise
򐂰 SQL Server Enterprise
򐂰 Teradata Enterprise
The ODBC Enterprise stage can only be used when a native parallel stage is not
available for the given source or target database.
When using Oracle, DB2, or Informix databases, use orchdbutil to import

design metadata.
Care must be taken to observe the data type mappings documented in

Chapter 13, “Database stage guidelines” on page 189, when designing a parallel
job with DataStage.
If possible, use a SQL where clause to limit the number of rows sent to a
DataStage job.
Avoid the use of database stored procedures on a per-row basis in a high-volume

data flow. For maximum scalability and parallel performance, it is best to
implement business rules natively using DataStage parallel components.

B.10 Troubleshooting and monitoring
Always test DS parallel jobs with a parallel configuration file
($APT_CONFIG_FILE) that has two or more nodes in its default pool.
Check the Director log for warnings, which might indicate an underlying problem
or data type conversion issue. All warnings and failures must be addressed (and
removed if possible) before deploying a DataStage job.
The environment variable $DS_PX_DEBUG can be used to capture all

generated OSH, error and warning messages from a running DS parallel job.
Set the environment variable $OSH_PRINT_SCHEMAS to capture actual

runtime schema to the Director log. Set $DS_PX_DEBUG if the schema record is
too large to capture in a Director log entry.
Enable $APT_DUMP_SCORE by default, and examine the job score by following

the guidelines outlined in Appendix E, “Understanding the parallel job score” on
page 401.

C
Appendix C. DataStage naming reference

Every name must be based on a three-part concept: Subject, Subject Modifier,
Class Word where the following frequently-used class words describe the object
type, or the function the object performs. In this appendix we provide a number of
tables, each for a specific category for easier access and understanding, with the
DataStage naming references. The first is Table C-1.
Table C-1 Project repository and components

Project Repository and Components
Development Dev_<proj>
Integration Test IT_<proj>
User Acceptance Test UAT_<proj>
Production Prod_<proj>
BuildOp BdOp<name>
Parallel External Function XFn<name>
Wrapper Wrap<name>

The next is in Table C-2.
Table C-2 Job names and properties

Job Names and Properties
Extract Job Src<job>
Load Load<job>
Sequence <job>_Seq
Parallel Shared Container <job>Psc
Server Shared Container <job>Ssc
Parameter <name>_parm
Table C-3 Sequencer

Sequencer
Job Activity Job
Routine Activity Rtn
Sequencer (All) SeqAll
Sequencer (Any) SeqAny
Notify Notify
Sequencer Links (messages) msg_

Table C-4 Links

Links (prefix with “lnk_”)
Reference (Lookup) Ref
Reject (Lookup, File, DB) Rej
Get (Shared Container) Get
Put (Shared Container) Put
Input In
Output Out
Delete Del
Insert Ins
Update Upd
Table C-5 Data store

Data Store
Database DB
Stored Procedure SP
Table Tbl
View View
Dimension Dim
Fact Fact
Source Src
Target Tgt
Appendix C. DataStage naming reference 393

Table C-6 Development and debug stages

Development / Debug stages
Column Generator CGen
Head Head
Peek Peek
Row Generator RGen
Sample Smpl
Tail Tail
Table C-7 File stages

File stages
Sequential File SF
Complex Flat File CFF
File Set FS
Parallel dataset DS
Lookup File Set LFS
External Source XSrc
External Target XTgt
Parallel SAS dataset SASd

Table C-8 Processing stages

Processing stages
Aggregator Agg
Change Apply ChAp
Change Capture ChCp
Copy Cp
Filter Filt
Funnel Funl
Join (Inner) InJn
Join (Left Outer) LOJn
Join (Right Outer) ROJn
Join (Full Outer) FOJn
Lookup Lkp
Merge Mrg
Modify Mod
Pivot Pivt
Remove Duplicates RmDp
SAS processing SASp
Sort Srt
Surrogate Key Generator SKey
Switch Swch
Table C-9 Transformer stage

Transformer stage
Transformer (native parallel) Tfm
BASIC Transformer (Server) BTfm
stage Variable SV
Appendix C. DataStage naming reference 395

Table C-10 Real-time stages

Real Time stages
RTI Input RTIi
RTI Output RTIo
XML Input XMLi
XML Output XMLo
XML Transformer XMLt
Table C-11 Restructure stages

Restructure stages
Column Export CExp
Column Import CImp

D
Appendix D. Example job template

This section summarizes the suggested job parameters for all DataStage jobs,
and presents them in the following tables. These might be defined in a job
template, which can be used by all developers for creating new parallel jobs.

Table D-1 Suggested environment variables for all jobs
$APT_CONFIG_FILE filepath Specifies the full path name to the DS parallel

configuration file. This variable must be included
in all job parameters so that it can be easily
changed at runtime.
$APT_DUMP_SCORE Outputs parallel score dump to the DataStage

job log, providing detailed information about
actual job flow including operators, processes,
and datasets. Extremely useful for
understanding how a job actually ran in the
environment.
$OSH_ECHO Includes a copy of the generated osh in the job’s

DataStage log
$APT_RECORD_COUNTS Outputs record counts to the DataStage job log

as each operator completes processing. The
count is per operator per partition.
This setting must be disabled by default, but part
of every job design so that it can be easily
enabled for debugging purposes.
$APT_PERFORMANCE_DATA $UNSET If set, specifies the directory to capture

advanced job runtime performance statistics.
$OSH_PRINT_SCHEMAS Outputs actual runtime metadata (schema) to

DataStage job log.
This setting must be disabled by default, but part
of every job design so that it can be easily
enabled for debugging purposes.
$APT_PM_SHOW_PIDS Places entries in DataStage job log showing

UNIX process ID (PID) for each process started
by a job. Does not report PIDs of DataStage
“phantom” processes started by Server shared
containers.
$APT_BUFFER_MAXIMUM_TIMEOUT Maximum buffer delay in seconds

Table D-2 Project_Plus environment variables
Name Type Prompt Default Value
$PROJECT_PLUS_DATASETS String Project + Dataset descriptor $PROJDEF

dir
$PROJECT_PLUS_LOGS String Project + Log dir $PROJDEF
$PROJECT_PLUS_PARAMS String Project + Parameter file dir $PROJDEF
$PROJECT_PLUS_SCHEMAS String Project + Schema dir $PROJDEF
$PROJECT_PLUS_SCRIPTS String Project + Scripts dir $PROJDEF
Table D-3 Staging environment variables

$STAGING_DIR String Staging directory $PROJDEF
$PROJECT_NAME String Project name $PROJDEF
$DEPLOY_PHASE String Deployment phase $PROJDEF
Table D-4 Job control parameters

JOB_NAME_parm String Job Name
RUN_ID_parm String Run ID
Appendix D. Example job template 399

E
Appendix E. Understanding the parallel

job score
DataStage parallel jobs are independent of the actual hardware and degree of
parallelism used to run them. The parallel configuration file provides a mapping
at runtime between the compiled job and the actual runtime infrastructure and
resources by defining logical processing nodes.
At runtime, the DS parallel framework uses the given job design and
configuration file to compose a job score that details the processes created,
degree of parallelism and node (server) assignments, and interconnects
(datasets) between them. Similar to the way a parallel database optimizer builds
a query plan, the parallel job score performs the following tasks:
򐂰 Identifies degree of parallelism and node assignments for each operator
򐂰 Details mappings between functional (stage/operator) and actual operating
system processes
򐂰 Includes operators automatically inserted at runtime:
– Buffer operators to prevent deadlocks and optimize data flow rates
between stages
– Sorts and Partitioners that have been automatically inserted to ensure
correct results

򐂰 Outlines connection topology (datasets) between adjacent operators and
persistent datasets
򐂰 Defines number of actual operating system processes
Where possible, multiple operators are combined in a single operating system

process to improve performance and optimize resource requirements.
E.1 Viewing the job score

When the environment variable APT_DUMP_SCORE is set, the job score is
output to the DataStage Director log. It is recommended that this setting be
enabled by default at the project level, as the job score offers invaluable data for
debugging and performance tuning, and the overhead to capture the score is
negligible.
As shown in Figure E-1, job score entries start with the phrase main_program:
This step has n datasets Two separate scores are written to the log for each
job run. The first score is from the license operator, not the actual job, and can be
ignored. The second score entry is the actual job score.
Figure E-1 Job score sample

E.2 Parallel job score components
The parallel job score is divided into two sections, as shown in Figure E-2 on
page 404:
򐂰 Datasets
Starts with the words main_program: This step has n datasets:
The first section details all datasets, including persistent (on disk) and virtual
(in memory, links between stages). Terminology in this section can be used to
identify the type of partitioning or collecting that was used between operators.
In this example, there are two virtual datasets.
򐂰 Operators:
Starts with the words It has n operators:
The second section details actual operators created to execute the job flow.
This includes:
– Sequential or Parallel operation, and the degree of parallelism per
operator.
– Node assignment for each operator. The actual node names correspond to
node names in the parallel configuration file. (in this example: “node1”,
“node2”, “node3”, “node4”).
Appendix E. Understanding the parallel job score 403

In Figure E-2, there are three operators, one running sequentially, two running in
parallel across four nodes, for a total of nine operating system process.
Figure E-2 Operators
The number of virtual datasets and the degree of parallelism determines the
amount of memory used by the inter-operator transport buffers. The memory
used by deadlock-prevention BufferOps can be calculated based on the number
of inserted BufferOps.

E.2.1 Job Score: Datasets
The parallel pipeline architecture passes data from upstream producers to
downstream consumers through in-memory virtual data sets. Figure E-3 depicts
an example.
Figure E-3 Data sets and operators in job score
Datasets are identified in the first section of the parallel job score, with each
dataset identified by its number (starting at zero). In this example, the first
dataset is identified as “ds0”, and the next “ds1”.
Producers and consumers might be either persistent (on disk) datasets or

parallel operators. Persistent datasets are identified by their dataset name.
Operators are identified by their operator number and name (see the lines
starting with "op0" and "op1"), , corresponding to the lower section of the job
score.
The degree of parallelism is identified in brackets after the operator name. For
example, operator zero (op0) is running sequentially, with one degree of
parallelism [1p]. Operator 1 (op1) is running in parallel with four degrees of
parallelism [4p].

In the dataset definition, the upstream producer is identified first, followed by a
notation to indicate the type of partitioning or collecting (if any), followed by the
downstream consumer. This is depicted in Figure E-4.
Producer
Partitioner Collector
Consumer
Figure E-4 Producer and consumer
The notation between producer and consumer is used to report the type of
partitioning or collecting (if any) that is applied. The partition type is associated
with the first term, collector type with the second. The symbol between the
partition name and collector name indicates the partition type and consumer. A
list of the symbols and their description is shown in Table E-1.
Table E-1 Partition types and consumers

> Sequential producer to Sequential consumer
<> Sequential producer to Parallel consumer
=> Parallel producer to Parallel consumer (SAME partitioning)
#> Parallel producer to Parallel consumer (repartitioned; not SAME)
>> Parallel producer to Sequential consumer
No producer or no consumer (typically, for persistent datasets)
Finally, if the Preserve Partitioning flag has been set for a particular dataset, the
notation “[pp]” appears in this section of the job score.

E.2.2 Job Score: Operators
In Figure E-5 we depict job score operators.
op0[1p] {(sequential APT_CombinedOperatorController:

(Row_Generator_0)
(inserted tsort operator {key={value=LastName},
key={value=FirstName}})
) on nodes (
node1[op0,p0]
)}
op1[4p] {(parallel inserted tsort operator {key={value=LastName}
key={value=FirstName}}(0))
on nodes (
node1[op2,p0]
node2[op2,p1]
node3[op2,p2]
node4[op2,p3]
)}
op2[4p] {(parallel buffer(0))
on nodes (
node1[op2,p0]
node2[op2,p1]
node3[op2,p2]
node4[op2,p3]
)}
Figure E-5 Job score operators
The lower portion of the parallel job score details the mapping between stages
and actual processes generated at runtime. For each operator, this includes (as
illustrated in the job score fragment) the following elements:
򐂰 Operator name (opn) numbered sequentially from zero (example “op0”)
򐂰 Degree of parallelism in brackets (example “[4p]”)
򐂰 Sequential or parallel execution mode
򐂰 Components of the operator, which have the following characteristics:
– Typically correspond to the user-specified stage name in the Designer
canvas
– Can include combined operators (APT_CombinedOperatorController),
which include logic from multiple stages in a single operator
– Can include framework-inserted operators such as Buffers, Sorts
– Can include composite operators (for example, Lookup)

Certain stages are composite operators. To the DataStage developer, a
composite operator appears to be a single stage on the design canvas. But
internally, a composite operator includes more than one function. This is depicted
in Figure E-6.
op2[1p] {(parallel APT_LUTCreateImpl in Lookup_3)

on nodes (
ecc3671[op2,p0]
)}
op3[4p] {(parallel buffer(0))
on nodes (
ecc3671[op3,p0]
ecc3672[op3,p1]
ecc3673[op3,p2]
ecc3674[op3,p3]
)}
op4[4p] {(parallel APT_CombinedOperatorController:
(APT_LUTProcessImpl in Lookup_3)
(APT_TransformOperatorImplV0S7_cpLookupTes
t1_Transformer_7 in Transformer_7)
(PeekNull)
) on nodes (
ecc3671[op4,p0]
ecc3672[op4,p1]
ecc3673[op4,p2]
ecc3674[op4,p3]
)}
Figure E-6 Composite operators
For example, Lookup is a composite operator. It is composite of the following

internal operators:
򐂰 APT_LUTCreateImpl
This operator reads the reference data into memory
򐂰 APT_LUTProcessImpl
This operator performs actual lookup processing after reference data has
been loaded
At runtime, each individual component of a composite operator is represented as

an individual operator in the job score, as shown in the following score fragment,
as depicted in Figure E-6.
Using this information together with the output from the $APT_PM_SHOW_PIDS
environment variable, you can evaluate the memory used by a lookup. Because
the entire structure needs to be loaded before actual lookup processing can
begin, you can also determine the delay associated with loading the lookup
structure.

In a similar way, a persistent dataset (shown in Figure E-7) defined to overwrite
an existing dataset of the same name has multiple entries in the job score to
perform the following tasks:
򐂰 Delete Data Files
򐂰 Delete Descriptor File
main_program: This step has 2 datasets:

ds0: {op1[1p] (parallel delete data files in delete temp.ds)
->eCollectAny
op2[1p] (sequential delete descriptor file in delete
temp.ds)}
ds1: {op0[1p] (sequential Row_Generator_0)
->
temp.ds}
It has 3 operators:
op0[1p] {(sequential Row_Generator_0)
on nodes (
node1[op0,p0]
)}
op1[1p] {(parallel delete data files in delete temp.ds)
on nodes (
node1[op1,p0]
)}
op2[1p] {(sequential delete descriptor file in delete temp.ds)
on nodes (
node1[op2,p0]
)}
It runs 3 processes on 1 node.
Figure E-7 Persistent dataset

F
Appendix F. Estimating the size of a

parallel dataset
For the advanced user, this Appendix provides a more accurate and detailed way
to estimate the size of a parallel dataset based on the internal storage
requirements for each data type. We have listed the data types and their sizes in
Table F-1 on page 412.

Table F-1 Data types and sizes
Data Type Size
Integers 4 bytes
Small Integer 2 bytes
Tiny Integer 1 byte
Big Integer 8 bytes
Decimal (precision+1)/2, rounded up
Float 8 bytes
VarChar(n) n + 4 bytes for non-NLS data

2n + 4 bytes for NLS data (internally stored as UTF-16)
Char(n) n bytes for non-NLS data

2n bytes for NLS data
Time 4 bytes
8 bytes with microsecond resolution
Date 4 bytes
Timestamp 8 bytes
12 bytes with microsecond resolution
For the overall record width, calculate and add the following values:
򐂰 (# nullable fields)/8 for null indicators
򐂰 one byte per column for field alignment (worst case is 3.5 bytes per field)
Using the internal DataStage parallel C++ libraries, the method

APT_Record::estimateFinalOutputSize() can give you an estimate for a given
record schema. APT_Transfer::getTransferBufferSize() can do this as well, if you
have a transfer that transfers all fields from input to output.

G
Appendix G. Environment variables

reference
In this Appendix we summarize the environment variables mentioned throughout
this document and have them listed in tables. These variables can be used on an
as-needed basis to tune the performance of a particular job flow, to assist in
debugging, or to change the default behavior of specific DataStage parallel
stages. An extensive list of environment variables is documented in the
DataStage Parallel Job Advanced Developers Guide.
The environment variable settings in this Appendix are only examples. Set values
that are optimal to your environment.
Table G-1 Job design environment variables

$APT_STRING_PADCHAR [char] Overrides the default pad character of 0x0 (ASCII

null) used when DS extends, or pads, a
variable-length string field to a fixed length (or a
fixed-length to a longer fixed-length).

Table G-2 Sequential File stage environment variables
$APT_EXPORT_FLUSH_ [nrows] Specifies how frequently (in rows) that the Sequential File
COUNT stage (export operator) flushes its internal buffer to disk.
Setting this value to a low number (such as 1) is useful for
real-time applications, but there is a small performance
penalty from increased I/O.
$APT_IMPORT_REJECT_ Setting this environment variable directs DataStage to
STRING_FIELD_OVERRUNS reject Sequential File records with strings longer than
their declared maximum column length. By default,
(DataStage v7.01 and later) imported string fields that exceed their maximum
declared length are truncated.
$APT_IMPEXP_ALLOW_ [set] When set, allows zero length null_field value with fixed
ZERO_LENGTH_FIXED_ length fields. Use this with care as poorly formatted data
NULL causes incorrect results. By default, a zero length
null_field value causes an error.
$APT_IMPORT_BUFFER_ [Kbytes] Defines size of I/O buffer for Sequential File reads
SIZE (imports) and writes (exports) respectively. Default is 128
$APT_EXPORT_BUFFER_ (128K), with a minimum of 8. Increasing these values on
SIZE heavily-loaded file servers can improve performance.
$APT_CONSISTENT_ [bytes] In certain disk array configurations, setting this variable to
BUFFERIO_SIZE a value equal to the read/write size in bytes can improve
performance of Sequential File import/export operations.
$APT_DELIMITED_READ [bytes] Specifies the number of bytes the Sequential File (import)
_SIZE stage reads-ahead to get the next delimiter. The default is
500 bytes, but this can be set as low as 2 bytes.
This setting must be set to a lower value when reading
from streaming inputs (for example, socket or FIFO) to
avoid blocking.
$APT_MAX_DELIMITED_ [bytes] By default, Sequential File (import) reads ahead 500
READ_SIZE bytes to get the next delimiter. If it is not found the
importer looks ahead 4*500=2000 (1500 more) bytes,
and so on (4X) up to 100,000 bytes.
This variable controls the upper bound ,which is, by
default, 100,000 bytes. When more than 500 bytes
read-ahead is desired, use this variable instead of
APT_DELIMITED_READ_SIZE.
$APT_IMPORT_PATTERN_ [set] When this environment variable is set (present in the
USES_FILESET environment) file pattern reads are done in parallel by
dynamically building a File Set header based on the list of
files that match the given expression. For disk
configurations with multiple controllers and disk, this
significantly improves file pattern reads.

Table G-3 Data set environment variables
$APT_PHYSICAL_DATASET [bytes] Specifies the size, in bytes, of the unit of data set I/O.
_BLOCK_SIZE Dataset segment files are written in chunks of this size. The
default is 128 KB (131,072)
$APT_OLD_BOUNDED_ [set] When this environment variable is set (present in the

LENGTH environment), Varchar columns is only stored using the
actual data length. This might improve I/O performance
(and reduce disk use) when processing a large number of
varchar columns with a large maximum length and
highly-variable data lengths.
Table G-4 DB2 environment variables

$INSTHOME [path] Specifies the DB2 install directory. This variable is

usually set in a user’s environment from .db2profile.
$APT_DB2INSTANCE_ [path] Used as a backup for specifying the DB2 installation

HOME directory (if $INSTHOME is undefined).
$APT_DBNAME [database] Specifies the name of the DB2 database for DB2/UDB
Enterprise stages if the Use Database Environment
Variable option is True. If $APT_DBNAME is not defined,
$DB2DBDFT is used to find the database name.
$APT_RDBMS_COMMIT_ [rows] Specifies the number of records to insert between

ROWS commits. The default value is 2000 per partition.
Can also be specified with the
“Row Commit Interval” stage
input property.
$DS_ENABLE_RESERVED_ Allows DataStage plug-in stages to handle DB2

CHAR_CONVERT databases that use the special characters # and $ in
column names.
Table G-5 Informix environment variables

$INFORMIXDIR [path] Specifies the Informix install directory.
$INFORMIXSQLHOSTS [filepath] Specifies the path to the Informix sqlhosts file.
$INFORMIXSERVER [name] Specifies the name of the Informix server

matching an entry in the sqlhosts file.
$APT_COMMIT_INTERVAL [rows] Specifies the commit interval in rows for Informix
HPL Loads. The default is 10000 per partiton.
Appendix G. Environment variables reference 415

Table G-6 Oracle environment variables
$ORACLE_HOME [path] Specifies installation directory for current Oracle instance.

Normally set in a user’s environment by Oracle scripts.
$ORACLE_SID [sid] Specifies the Oracle service name, corresponding to a

TNSNAMES entry.
$APT_ORAUPSERT_ [num] These two environment variables work together to specify

COMMIT_ROW_INTERVAL [seconds] how often target rows are committed for target Oracle
stages with Upsert method.
$APT_ORAUPSERT_
COMMIT_TIME_INTERVAL Commits are made whenever the time interval period has
passed or the row interval is reached, whichever comes
first. By default, commits are made every two seconds or
5000 rows per partition.
$APT_ORACLE_LOAD_ [SQL* Specifies Oracle SQL*Loader options used in a target

OPTIONS Loader Oracle stage with Load method. By default, this is set to
options] OPTIONS(DIRECT=TRUE, PARALLEL=TRUE)
$APT_ORACLE_LOAD_ [char] Specifies a field delimiter for target Oracle stages using the
DELIMITED Load method. Setting this variable makes it possible to load
fields with trailing or leading blank characters.
(DataStage 7.01 and later)
$APT_ORA_IGNORE_ When set, a target Oracle stage with Load method limits the
CONFIG_FILE_ number of players to the number of datafiles in the table’s
PARALLELISM table space.
$APT_ORA_WRITE_FILES [filepath] Useful in debugging Oracle SQL*Loader issues. When set,

the output of a Target Oracle stage with Load method is
written to files instead of invoking the Oracle SQL*Loader.
The filepath specified by this environment variable specifies
the file with the SQL*Loader commands.
$DS_ENABLE_RESERVED Allows DataStage plug-in stages to handle Oracle

_CHAR_CONVERT databases that use the special characters # and $ in column
names.

Table G-7 Teradata environment variables
$APT_TERA_SYNC_ [name] Starting with V7, specifies the database used for the
DATABASE terasync table.
$APT_TERA_SYNC_USER [user] Starting with V7, specifies the user that creates and
writes to the terasync table.
$APT_TER_SYNC_ [password] Specifies the password for the user identified by

PASSWORD $APT_TERA_SYNC_USER.
$APT_TERA_64K_BUFFERS Enables 64 K buffer transfers (32 K is the default).

Might improve performance depending on network
configuration.
$APT_TERA_NO_ERR_ When set, this environment variable might assist in

CLEANUP job debugging by preventing the removal of error
tables and partially written target table.
This environment variable is not recommended for
general use.
$APT_TERA_NO_PERM_ Disables permission checking on Teradata system

CHECKS tables that must be readable during the TeraData
Enterprise load process. This can be used to improve
the startup time of the load.
Table G-8 Netezza environment variables

$NETEZZA [path] Specifies the Nezza home directory
$NZ_ODBC_INI_PATH [filepath] Points to the location of the .odbc.ini file. This

is required for ODBC connectivity on UNIX
systems.
$APT_DEBUG_MODULE_ odbcstmt, odbcenv, Prints debug messages from a specific

NAMES nzetwriteop, nzutils, DataStage module. Useful for debugging
nzwriterep, nzetsubop Netezza errors

Table G-9 Job monitoring environment variables
$APT_MONITOR_TIME [seconds] In V7 and later, specifies the time interval (in seconds)
for generating job monitor information at runtime. To
enable size-based job monitoring, unset this
environment variable, and set $APT_MONITOR_SIZE.
$APT_MONITOR_SIZE [rows] Determines the minimum number of records the job

monitor reports. The default of 5000 records is usually
too small. To minimize the number of messages during
large job runs, set this to a higher value (for example,
1000000).
$APT_NO_JOBMON Disables job monitoring completely. In rare instances,

this might improve performance. In general, this can
only be set on a per-job basis when attempting to
resolve performance bottlenecks.
$APT_RECORD_COUNTS Prints record counts in the job log as each operator

completes processing. The count is per operator per
partition.

Table G-10 Performance-tuning environment variables
$APT_BUFFER_MAXIMUM_ 41903040 Specifies the maximum amount of virtual memory in

MEMORY (example) bytes used per buffer per partition. If not set, the default
is 3 MB (3145728). Setting this value higher uses more
memory, depending on the job flow, but might improve
performance.
$APT_BUFFER_FREE_RUN 1000 Specifies how much of the available in-memory buffer

(example) to consume before the buffer offers resistance to any
new data being written to it. If not set, the default is 0.5
(50% of $APT_BUFFER_MAXIMUM_MEMORY).
If this value is greater than 1, the buffer operator reads
$APT_BUFFER_FREE_RUN *
$APT_BUFFER_MAXIMIMUM_MEMORY before
offering resistance to new data.
When this setting is greater than 1, buffer operators
spool data to disk (by default scratch disk) after the
$APT_BUFFER_MAXIMUM_MEMORY threshold. The
maximum disk required is
$APT_BUFFER_FREE_RUN * # of buffers *
$APT_BUFFER_MAXIMUM_MEMORY
$APT_PERFORMANCE_ directory Enables capture of detailed, per-process performance

DATA data in an XML file in the specified directory. Unset this
environment variable to disable.
$TMPDIR [path] Defaults to /tmp. Used for miscellaneous internal

temporary data including FIFO queues and
Transformer temporary storage.
As a minor optimization, might be best set to a
filesystem outside of the DataStage install directory.

Table G-11 Job flow debugging environment variables
$OSH_PRINT_SCHEMAS Outputs the actual schema definitions used by the

DataStage parallel framework at runtime in the
DataStage log. This can be useful when determining if
the actual runtime schema matches the expected job
design table definitions.
$APT_DISABLE_ Disables operator combination for all stages in a job,

COMBINATION forcing each parallel operator into a separate process.
Though not normally needed in a job flow, this setting
The Advanced stage might help when debugging a job flow or investigating
Properties editor in DataStage performance by isolating individual operators to
Designer v7.1 and later allows separate processes.
combination to be enabled
and disabled for on a Note that disabling operator combination generates
per-stage basis. more UNIX processes, and requires more system
resources (and memory). Disabling operator
combination also disables internal optimizations for job
efficiency and run-times.
$APT_PM_PLAYER_TIMING Prints detailed information in the job log for each

operator, including CPU use and elapsed processing
time.
$APT_PM_PLAYER_ Prints detailed information in the job log for each

MEMORY operator when allocating additional heap memory.
$APT_BUFFERING_POLICY FORCE Forces an internal buffer operator to be placed between

Setting every operator. Normally, the DataStage framework
$APT_BUFFERING_POLICY inserts buffer operators into a job flow at runtime to
=FORCE is not avoid deadlocks and improve performance.
recommended for
production job runs. Using $APT_BUFFERING_POLICY=FORCE in
combination with $APT_BUFFER_FREE_RUN
effectively isolates each operator from slowing
upstream production. Using the job monitor
performance statistics, this can identify which part of a
job flow is impacting overall performance.
$DS_PX_DEBUG Set this environment variable to capture copies of the

job score, generated osh, and internal parallel
framework log messages in a directory corresponding
to the job name. This directory is created in the
Debugging sub-directory of the Project home directory
on the DataStage server.

$APT_PM_STARTUP_ This environment variable should not normally need to
CONCURRENCY be set. When trying to start large jobs on heavily-loaded
servers, lowering this number limits the number of
processes that are simultaneously created when a job is
started.
$APT_PM_NODE_TIMEOUT [seconds] For heavily -loaded MPP or clustered environments, this

variable determines the number of seconds the
conductor node waits for a successful startup from each
section leader. The default is 30 seconds.

H
Appendix H. DataStage data types

The DataStage Designer represents column data types using SQL notation.
Each SQL data type maps to an underlying data type in the DS parallel
framework. The internal parallel data types are used in schema files and are
displayed when viewing generated OSH or viewing the output from
$OSH_PRINT_SCHEMAS.

H.1 Parallel data types
In Table H-1 we summarize the underlying parallel data types of DataStage.
Table H-1 DataStage parallel data types

SQL Type Internal Type Size Description
Date date 4 bytes Date with month, day, and year
Decimal, decimal (roundup(p)+ Packed decimal, compatible with IBM packed

Numeric 1)/2 decimal format.
Float, sfloat 4 bytes IEEE single-precision (32-bit) floating point

Real value
Double dfloat 8 bytes IEEE double-precision (64-bit) floating point

value
TinyInt int8, uint8 1 byte Signed or unsigned integer of 8 bits

(Specify unsigned Extended option for
unsigned)
SmallInt int16, uint16 2 bytes Signed or unsigned integer of 16 bits (Specify

unsigned Extended option for unsigned)
Integer int32, unit32 4 bytes Signed or unsigned integer of 32 bits (Specify

BigInta int64, unit64 8 bytes Signed or unsigned integer of 64 bits (Specify

Binary, Bit, raw 1 byte per Untyped collection, consisting of a fixed or

LongVarBinary, character variable number of contiguous bytes and an
VarBinary optional alignment value
Unknown, Char, string 1 byte per ASCII character string of fixed or variable
LongVarChar, character length (Unicode Extended option NOT
VarChar selected)
NChar, ustring multiple bytes ASCII character string of fixed or variable

NVarChar, per character length (Unicode Extended option NOT
LongNVarChar selected)
Char, ustring multiple bytes ASCII character string of fixed or variable

LongVarChar, per character length (Unicode Extended option IS selected)
VarChar
Time time 5 bytes Time of day, with resolution to seconds
Time time 5 bytes Time of day, with resolution of microseconds

(microseconds) (Specify microseconds Extended option)

Timestamp timestamp 9 bytes Single field containing both date and time value
with resolution to seconds.
Timestamp timestamp 9 bytes Single field containing both date and time value
(microseconds) with resolution to microseconds. (Specify
microseconds Extended option)
a. BigInt values map to long long integers on all supported platforms except Tru64 where they map to longer
integer values.
Strings and Ustrings

If NLS is enabled on your DataStage server, parallel jobs support two types of
underlying character data types:
򐂰 Strings
򐂰 Ustrings
String data represents unmapped bytes, ustring data represents full Unicode
(UTF-16) data.
The Char, VarChar, and LongVarChar SQL types relate to underlying string types
where each character is 8-bits and does not require mapping because it
represents an ASCII character. You can, however, specify that these data types
are extended, in which case they are taken as ustrings and require mapping.
(They are specified as such by selecting the “Extended” check box for the column
in the Edit Meta Data dialog box.) An Extended field appears in the columns grid,
and extended Char, VarChar, or LongVarChar columns have Unicode in this field.
The NChar, NVarChar, and LongNVarChar types relate to underlying ustring
types, so do not need to be explicitly extended.
Appendix H. DataStage data types 425

Default and explicit type conversions
DataStage provides a number of default conversions and conversion functions
when mapping from a source to a target parallel data type. Default type
conversions take place across the stage output mappings of any parallel stage.
Figure H-1 summarizes Data Type conversions.
Target Field
Source d = There is a default type conversion from source field type to destination field type.
Field e = You can use a Modify or a Transformer conversion function to explicitly convert from
the source field type to the destination field type.
A blank cell indicates that no conversion is provided.
timestamp
unstring
decimal
uint64
uint16
uint32
dfloat
string
sfloat
uint8
int16
int64
int32
date
time
int8
raw
int8 de de de
uint8
int16 de de de
uint16 de de
int32 de de de
uint32
int64 de
uint64
sfloat de
dfloat de de de de
decimal de de de de de de de
string de de de de de
unstring de de de de de
raw
date
time de
timestamp
Figure H-1 Data type conversions

The conversion of numeric data types can result in a loss of precision and cause
incorrect results, depending on the source and result data types. In these
instances, the parallel framework displays a warning message in the job log.

When converting from variable-length to fixed-length strings, the parallel framework
pads the remaining length with NULL (ASCII zero) characters by default.
򐂰 The environment variable APT_STRING_PADCHAR can be used to change
the default pad character from an ASCII NULL (0x0) to another character; for
example, an ASCII space (0x20) or a Unicode space (U+0020). When
entering a space for the value of APT_STRING_PADCHAR do note enclose
the space character in quotes.
򐂰 As an alternate solution, the PadString Transformer function can be used to
pad a variable-length (Varchar) string to a specified length using a specified
pad character. Note that PadString does not work with fixed-length (CHAR)
string types. You must first convert a Char string type to a Varchar type before
using PadString.
򐂰 Certain stages (for example, Sequential File and DB2/UDB Enterprise
targets) allow the pad character to be specified in their stage or column
definition properties. When used in these stages, the specified pad character
overrides the default for that stage only.
H.2 Null handling

The DataStage parallel framework represents nulls in two ways:
򐂰 It allocates a single bit to mark a field as null. This type of representation is
called an out-of-band null.
򐂰 It designates a specific field value to indicate a null (for example a numeric
field’s most negative possible value). This type of representation is called an
in-band null. In-band null representation can be disadvantageous because
you must reserve a field value for nulls, and this value cannot be treated as
valid data elsewhere.
The Transformer and Modify stages can change a null representation from an
out-of-band null to an in-band null and from an in-band null to an out-of-band null.
Note: When processing nullable columns in a Transformer stage, care must

be taken to avoid data rejects. See 9.1.1, “Transformer NULL handling and
reject link” on page 140.
When reading from dataset and database sources with nullable columns, the
DataStage parallel framework uses the internal, out-of-band null representation
for NULL values.

When reading from or writing to Sequential Files or File Sets, the in-band (value)
must be explicitly defined in the extended column attributes for each Nullable
column, as shown in Figure H-2.
Figure H-2 Extended column metadata (Nullable properties)

The Table Definition of a stage’s input or output data set can contain columns
defined to support out-of-band nulls (Nullable attribute is checked). In Table H-2
we list the rules for handling nullable fields when a stage takes a dataset as input
or writes to a dataset as output.
Table H-2 Table definitions

Source Field Destination Field Result
not Nullable not Nullable Source value propagates to destination.
Nullable Nullable Source value or null propagates.
not Nullable Nullable Source value propagates;. Destination value is

never null.
Nullable not Nullable If the source value is not null, the source value
propagates.
If the source value is null, a fatal error occurs.
H.3 Runtime column propagation

Runtime column propagation (RCP) allows job designs to accommodate
additional columns beyond those defined by the job developer. Before a
DataStage developer can use RCP, it must be enabled at the project level
through the administrator client.
Using RCP judiciously in a job design facilitates re-usable job designs based on
input metadata, rather than using a large number of jobs with hard-coded table
definitions to perform the same tasks. Certain stages, for example the Sequential
File stage, allow their runtime schema to be parameterized, further extending
re-use through RCP.
Furthermore, RCP facilitates re-use through parallel shared containers. Using

RCP, only the columns explicitly referenced in the shared container logic need to
be defined, the remaining columns pass through at runtime, as long as each
stage in the shared container has RCP enabled on their stage Output properties.

Related publications
The publications listed in this section are considered particularly suitable for a
more detailed discussion of the topics covered in this book.
IBM Redbooks
For information about ordering these publications, see “How to get Redbooks” on
page 432. Note that some of the documents referenced here might be available
in softcopy only.
򐂰 IBM WebSphere QualityStage Methodologies, Standardization, and
Matching, SG24-7546
Other publications
These publications are also relevant as further information sources:
򐂰 IBM Information Server 8.1 Planning, Installation and Configuration Guide,
GC19-1048
򐂰 IBM Information Server Introduction, GC19-1049
Online resources
These Web sites are also relevant as further information sources:
򐂰 IBM Information Server information center
http://publib.boulder.ibm.com/infocenter/iisinfsv/v8r0/index.jsp
򐂰 IBM Information Server Quick Start Guide
http://www-01.ibm.com/support/docview.wss?uid=swg27009391&aid=1

How to get Redbooks
You can search for, view, or download Redbooks, Redpapers, Technotes, draft
publications and Additional materials, as well as order hardcopy Redbooks
publications, at this Web site:
ibm.com/redbooks
Help from IBM

IBM Support and downloads
ibm.com/support
IBM Global Services

ibm.com/services

InfoSphere DataStage: Parallel Framework Standard Practices
(0.5” spine)
0.475”<->0.875”
250 <-> 459 pages
Back cover ®
InfoSphere DataStage
Parallel Framework
Standard Practices ®
Develop highly In this IBM Redbooks publication, we present guidelines for

efficient and the development of highly efficient and scalable information INTERNATIONAL
scalable information integration applications with InfoSphere DataStage (DS) TECHNICAL
integration parallel jobs. SUPPORT
applications InfoSphere DataStage is at the core of IBM Information
ORGANIZATION
Server, providing components that yield a high degree of
Investigate, design, freedom. For any particular problem there might be multiple
and develop data solutions, which tend to be influenced by personal BUILDING TECHNICAL
flow jobs preferences, background, and previous experience. All too INFORMATION BASED ON
often, those solutions yield less than optimal, and PRACTICAL EXPERIENCE
non-scalable, implementations.
Get guidelines for
cost effective This book includes a comprehensive detailed description of IBM Redbooks are developed by
performance the components available, and descriptions on how to use the IBM International Technical
them to obtain scalable and efficient solutions, for both batch Support Organization. Experts
and real-time scenarios. from IBM, Customers and
Partners from around the world
The advice provided in this document is the result of the create timely technical
combined proven experience from a number of expert information based on realistic
practitioners in the field of high performance information scenarios. Specific
recommendations are provided
integration, evolved over several years. to help you implement IT
This book is intended for IT architects, Information solutions more effectively in
Management specialists, and Information Integration your environment.
specialists responsible for delivering cost-effective IBM
InfoSphere DataStage performance on all platforms.
For more information:
ibm.com/redbooks
SG24-7830-00 0738434477

Parallel Standard Practices DS

Uploaded by

Copyright:

Available Formats

Parallel Standard Practices DS

Uploaded by

Copyright:

Available Formats

Front cover

Investigate, design, and develop

Get guidelines for cost

InfoSphere DataStage: Parallel Framework

First Edition (September 2010)

© Copyright International Business Machines Corporation 2010. All rights reserved.

Chapter 1. Data integration with Information Server and DataStage . . . . . 1

Chapter 2. Data integration overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

© Copyright IBM Corp. 2010. All rights reserved. iii

Chapter 5. Development guidelines. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

Chapter 6. Partitioning and collecting. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

iv InfoSphere DataStage: Parallel Framework Standard Practices

Chapter 7. Sorting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

Chapter 8. File Stage usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

Chapter 9. Transformation languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139

Chapter 10. Combining data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

Chapter 11. Restructuring data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

Chapter 12. Performance tuning job designs . . . . . . . . . . . . . . . . . . . . . . 173

Chapter 13. Database stage guidelines. . . . . . . . . . . . . . . . . . . . . . . . . . . 189

vi InfoSphere DataStage: Parallel Framework Standard Practices

Chapter 15. Batch data flow design. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259

viii InfoSphere DataStage: Parallel Framework Standard Practices

Chapter 16. Real-time data flow design. . . . . . . . . . . . . . . . . . . . . . . . . . . 293

Appendix A. Runtime topologies for distributed transaction jobs. . . . . 375

Appendix B. Standard practices summary . . . . . . . . . . . . . . . . . . . . . . . . 381

Appendix C. DataStage naming reference . . . . . . . . . . . . . . . . . . . . . . . . 391

x InfoSphere DataStage: Parallel Framework Standard Practices

Appendix E. Understanding the parallel job score . . . . . . . . . . . . . . . . . 401

Appendix F. Estimating the size of a parallel dataset . . . . . . . . . . . . . . . 411

Appendix G. Environment variables reference. . . . . . . . . . . . . . . . . . . . . 413

Appendix H. DataStage data types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423

Related publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431

© Copyright IBM Corp. 2010. All rights reserved. xiii

AIX® Information Agenda™ MVS™

The following terms are trademarks of other companies:

xiv InfoSphere DataStage: Parallel Framework Standard Practices

In this IBM® Redbooks® publication, we present guidelines for the development

InfoSphere DataStage is at the core of IBM Information Server, providing

This book includes a comprehensive detailed description of the components

This book is intended for IT architects, Information Management specialists, and

This book is organized as follows:

© Copyright IBM Corp. 2010. All rights reserved. xv

xvi InfoSphere DataStage: Parallel Framework Standard Practices

The team who wrote this book

Julius Lerm is a Technical Architect, and a member of the Center of Excellence

From the International Technical Support Organization

xviii InfoSphere DataStage: Parallel Framework Standard Practices

We want our books to be as helpful as possible. Send us your comments about

xx InfoSphere DataStage: Parallel Framework Standard Practices

Chapter 1. Data integration with

IBM InfoSphere Information Server is a software platform that helps

© Copyright IBM Corp. 2010. All rights reserved. 1

DataStage can manage data arriving in real-time as well as data received on a

Along with these key components, establishing consistent development

But first, to aid in understanding, we provide a brief overview of the InfoSphere

2 InfoSphere DataStage: Parallel Framework Standard Practices

DataStage 7.X consisted of a two-tier infrastructure, with clients connected

1.1.1 Architecture and information tiers

Chapter 1. Data integration with Information Server and DataStage 3

Services Tier Engine Tier

Services for IBM IBM Information

Figure 1-1 Information Server tiers

In this document we focus on aspects related to parallel job development, which

4 InfoSphere DataStage: Parallel Framework Standard Practices

Figure 1-2 IBM InfoSphere Services overview

Chapter 1. Data integration with Information Server and DataStage 5

Table 1-1 Services description

1.3 Center of Excellence for Data Integration (CEDI)

6 InfoSphere DataStage: Parallel Framework Standard Practices

Installation and Configuration

Strategic Test & Monitor &

Requirements Definition, Administration, Management, and