SQL Server To SQL Server PDW: Migration Guide (AU3)

SQL Server
to SQL
Server PDW
Migration Guide
(AU3)
Contents
4 Summary Statement
4 Introduction
4 SQL Server Family of Products
6 Differences between SMP and MPP
8 PDW Software Architecture
10 PDW Community
10 Migration Planning
11 Determine Candidates for Migration
11 Migration Considerations
12 Migration Approaches
13 SQL Server to PDW Migration

13 Main Migration Steps
13 PDW Migration Advisor
14 APS Migration Utility
14 Table Geometry
20 Optimization within PDW
Microsoft SQL Server to SQL Server PDW Migration Guide (AU3)

29 Converting Database Schema Objects
29 Type Mapping
36 Syntax Differences
51 Conclusion
51 For More Information
51 Feedback
52 Appendix

2014 Microsoft Corporation. All rights
reserved. This document is provided as-is.
Information and views expressed in this
document, including URL and other Internet
Web site references, may change without
notice. You bear the risk of using it. This
document does not provide you with any legal
rights to any intellectual property in any
Microsoft product. You may copy and use this
document for your internal, reference
purposes. You may modify this document for
your internal, reference purposes.
In this migration guide you will learn the differences between the
Summary SQL Server and Parallel Data Warehouse database platforms, which
runs on the Analytics Platform System appliance, and the steps
Statement necessary to convert a SQL Server database to Parallel Data

Warehouse.
This guide has been updated to reflect the T-SQL improvements

contained within Version 2, Appliance Update 3 (AU3)
Migrating a data warehouse application from Microsoft SQL Server

Introduction SMP to the Microsoft SQL Server Parallel Data Warehouse (PDW)
region within the Microsoft Analytics Platform System (APS) appliance,
provides many benefits, including improved query performance,
scalability and the simplest integration to Hadoop.
This white paper explores the remediation activities and provides

guidance to overcome technical differences between the two database
platforms in order to simplify migration projects. It describes the
implementation differences of database objects, SQL dialects, and
procedural code between the two platforms.
SQL Server Microsoft offers a wide range of SQL Server products to accommodate
the different business needs and technical requirements to provide a
Family of robust and scalable enterprise-wide solution.
Products
Compact Edition
Microsoft SQL Server Compact 4.0 is a compact database ideal for
embedding in desktop and web applications. SQL Server Compact 4.0
gives developers a common programming model with other SQL Server
editions for developing both native and managed applications. SQL
Server Compact provides relational database functionality in a small
footprint.
Express Edition
SQL Server 2014 Express Edition is available for free from Microsoft and
provides a powerful database engine ideal for embedded applications or
for redistribution with other solutions. Independent software vendors
use it to build desktop and data-driven applications. If you need more
advanced database features and support for greater than 10 GB
databases, SQL Server Express is fully compatible with other editions of
SQL Server can be seamlessly upgraded to enterprise versions of SQL
Microsoft SQL Server to SQL Server PDW Migration Guide (AU3) 4

Server without requiring remediation to overcome changes in
functionality.
Standard Edition
SQL Server 2014 Standard edition is a robust data management and
business intelligence database for departments and small workgroups to
support a wide variety of applications. SQL Server Standard Edition also
supports common development tools for on premise and cloud.
Enabling effective database management with minimal IT resources. SQL
Server Standard Edition is compatible with other editions.
Web Edition
SQL Server 2014 Web edition is a low total-cost-of-ownership option for
Web hosters and Web VAPs to provide scalability, affordability, and
manageability capabilities for small to large scale Web properties.
Business Intelligence Edition

SQL Server 2014 Business Intelligence edition delivers a comprehensive
platform empowering organizations to build and deploy secure, scalable
and manageable BI solutions. It offers exciting functionality such as
browser based data exploration and visualization; powerful data mash-
up capabilities, and enhanced integration management.
Enterprise Edition
SQL Server 2014 Enterprise edition delivers comprehensive high-end
datacenter capabilities with blazing-fast performance, unlimited
virtualization, and end-to-end business intelligence. Enabling high
service levels for mission-critical workloads and end user access to data
insights.
Parallel Data Warehouse

While the other versions of SQL Server mentioned here have a
symmetric multi-processing (SMP) architecture, SQL Server Parallel Data
Warehouse has a massively parallel processing (MPP) architecture, in the
form of a data warehousing appliance, designed and built for to manage
high volumes of relational data (with up to 100x performance gains) and
providing the simplest integration to Hadoop. This addition of SQL
Server only runs on the Microsoft Analytics Platform System appliance;
available from HP, Dell and Quanta.

Differences As data volumes grow and the number of users increase, we find the
traditional architecture of running your data warehouse on a single SMP
between server insufficient to meet the business requirements. To overcome

these limitations and scale and perform beyond your existing
SMP and
deployment, Parallel Data Warehouse brings Massively Parallel
Processing (MPP) to the world of SQL Server. Essentially parallelizing and
MPP
distributing the processing across multiple SMP compute nodes. SQL
Server Parallel Data Warehouse is only available as part of Microsofts
Analytics Platform System (APS) appliance.
Before diving into the PDW architecture, let us first understand the
differences between the architectures mentioned above.
Symmetric Multi-Processing (SMP)

Symmetric Multi-Processing (SMP) is the primary parallel architecture
employed in servers. An SMP architecture is a tightly coupled
multiprocessor system, where processors share a single copy of the
operating system (OS) and resources that often include a common bus,
memory and an I/O system.
Think of this as a typical single server, multi-core with locally attached

storage, running the Microsoft Windows OS.
Massively Parallel Processing (MPP)

Massively Parallel Processing (MPP) is the coordinated processing of a
single task by multiple processers, each working on a different part of
the task. With each processor using its own operating system (OS) and
memory. MPP processors communicate between each other using some
form of messaging interface via an interconnect. The setup for MPP is
more complicated than SMP and one approach to simplify this while
providing equal amounts of resources between processors is the
Shared-Nothing architecture. This approach is the one Parallel Data
Warehouse is based upon.
Shared Nothing Architecture

The term shared nothing architecture was coined by Michael
Stonebraker (1986) to describe a multiprocessor database management
system in which neither memory nor disk storage is shared among the
processors.
For a database which follows the shared-nothing architecture, each

processor has its own set of disks. Data is horizontally partitioned
across nodes, such that each node has a subset of the rows from each
table in the database. Each node is then responsible for processing only
the rows on its own disks. Such architectures are especially well suited to
data warehouse workloads, where large fact tables can be distributed
across the nodes.
In addition, every node maintains its own lock table and buffer pool,
eliminating the need for complicated locking and software or hardware
consistency mechanisms. Because shared nothing does not typically
have a severe bus or resource contention it can be made to scale to
hundreds or even thousands of machines.

PDW
Software
Architecture
Admin Console
The Admin Console is a web application that visualises the
appliance state, health, and performance information.
MPP Engine
The MPP Engine is the brain of the SQL Server Parallel Data Warehouse
(PDW) and delivers the Massively Parallel Processing (MPP) capabilities
by doing the following:
Generates the parallel query execution plan and coordinates the

parallel query execution across the compute nodes.
Stores and coordinates metadata and configuration data for all of

the databases.
Manages SQL Server PDW database authentication and

authorization.
Tracks hardware and software status.

Data Movement Service (DMS)
Transfers data between the Computer nodes and the Control node.
Processes query operations that require transferring data among

the nodes.
Improves query performance by optimizing data transfer speeds.
Data is loaded in parallel directly from the loading server to the

Compute nodes via the DMS.
DMS transfers data from each Compute node directly to the

backup server.
Using PolyBase, DMS transfers data to and from an external

Hadoop cluster, or the HDInsight Region on the appliance
SQL Server Databases

Each Compute node runs an instance of SQL Server to process
queries and manage user data.
The Shell database manages the metadata for all distributed user
databases.
TempDB contains the metadata for all user temporary tables across
the appliance.
Master is the master table for SQL Server on the Control node
Configuration Tool
The Configuration Tool (dwconfig.exe) is used by appliance
administrators to configure the Analytics Platform System.
Domain Controller
Performs authentication among the Analytics Platform System
nodes, and manages the authentication of SQL Server PDW
Windows Authentication logins
Windows Domain Name Service (DNS) resolves domain names to

IP addresses for the Analytics Platform System appliance.
Windows Deployment Service

Windows Deployment Service (WDS) deploys the Windows Server
operating system onto the appliance. It is deployed on every host
and virtual machine across the appliance.
The DHCP service creates IP addresses so that the hosts within the
appliance domain can join the appliance network without having a
pre-configured IP address.

Virtual Machine Manager
Analytics Platform System uses virtualization to achieve high
availability. The Virtual Machine Manager hosts System Center to
deploy the operating system on the physical hosts.
Windows Server Update Services (WSUS) provides the ability to

apply or remove Windows Updates across all of the hosts and
virtual machines.
PDW Microsoft APS Documentation and Client Tools

The Microsoft Analytics Platform System product documentation and
Community client tools can be obtained via the following locations:
Microsoft Download Center
http://www.microsoft.com/en-us/download/details.aspx?id=45294
Microsoft APS Instructor-Led Training

A number of different Microsoft Partners offer instructor-led training
about Microsoft APS. Contact your Microsoft Account Manager to
obtain a list of these partners within your region.
An online course is available on the Microsoft Virtual Academy at:
http://www.microsoftvirtualacademy.com/training-courses/big-
data-with-the-microsoft-analytics-platform-system
Microsoft Premier Support for APS

Microsoft Premier Support for APS is a service offered by Microsoft
providing support through accredited appliance Microsoft Support
Engineers, remote managed upgrades and update assistance.
Before you begin the migration of an existing SQL Server data

Migration warehouse to SQL Server Parallel Data Warehouse (PDW) it is
recommended that you first plan what it is you want to migrate and
Planning make a decision on the migration approach.

Determine Migration of a data warehouse provides the perfect opportunity to
remove data, objects and processes which are no longer utilized.
Candidates Therefore the first step for migration planning is to identify and
determine which objects and processes you wish to migrate.
for Identify candidate databases for migration
Migration Primary Data Warehouses
Primary Data Marts
Staging Databases
Archive Databases
Identify candidate objects for migration
Tables
Views
Stored Procedures
Identify candidate data for migration
Archive non-required data
Migration After determining candidates for migration, you should also consider
the following areas and determine how you wish to manage them as
Considerations part of the migration. For example, you may be planning to migrate
multiple legacy SQL Server data marts onto a single SQL Server Parallel
Data Warehouse, doing so may increase the importance of the APS
appliance and therefore require an increase in availability and disaster
recovery.
Mission Critical
High-Availability & Disaster Recover
Security & Data Encryption
Database Design
Number of Users & Concurrency
Design Complexities (ELT/ETL, UDTs, UDFs, Physical Storage)

Data Loads (Volume, Frequency, Latency)
Capacity Planning (current and future storage requirements)
Backup & Restore
Migration There are many different options to migrating data and applications
from an existing SQL Server system to a SQL Server Parallel Data
Approaches Warehouse, each of which provides different costs and benefits.
The three most widely adopted migration approaches are:
1:1 Migration The existing data model and schema is moved

directly to the SQL Server PDW with little or no change to the
model. The benefits being the increasing performance, availability,
scalability and the decreasing cost of ownership (TCO) of the
platform. Sometimes referred to as Re-Hosting or the Forklift
approach.
Redesign The data model is redesigned and the application is re-

architected following the SQL Server PDW best practices. This can
help reduce overall complexity and enable the ability for the data
warehouse to be more flexible and answer any question at any
time.
Evolution The evolutionary approach is a combination of the

above two, allowing for the new target platform to be up and
running quickly and over time deliver increasing benefits and
capabilities by redesigning.

Determining the overall goal and reason for migration will provide the
means to determine which of the 3 approaches is best suited.
Migration Drivers Migration Approach
Performance 1:1 Migration or Redesign or Evolution
Availability 1:1 Migration or Evolution
Scalability 1:1 Migration or Evolution
Single View of the Business Redesign or Evolution
Business Questions (Complexity) Redesign or Evolution
Cost of Ownership (TCO) 1:1 Migration
This section provides an overview of SQL Server to PDW migration. It
SQL Server discusses main migration steps, PDW Migration Advisor, table geometry,
and optimization within PDW.
to PDW
Migration
Main This white paper provides you guidance in migrating the objects and
highlights areas which are not supported within SQL Server PDW. To aid
Migration with identifying the migration areas to focus on, you can make use of
the PDW Migration Advisor.
Steps Once all objects have been migrated, the next step would be to migrate
any integration or ETL jobs.
PDW The PDW Migration Advisor (PDWMA) is a tool that can be used to
inspect your Microsoft SQL Server database schemas in order to display
Migration the changes that need to be made in order to migrate the database to
Microsoft SQL Server PDW.
Advisor

PDWMA is able to review the DDL and DML code structured and
perform a series of validation checks to determine compatibility with
PDW. This will significantly reduce the investigation efforts required in
migration scoping projects. This whitepaper provides guidance in order
to overcome technical issues that may be encountered in the database
migration project.
A representative from the Microsoft technical pre-sales team will be

happy to run the PDWMA for you please connect with your Microsoft
representative to run this tool/free service.
APS The APS Migration Utility is a tool that can be used to automatically
migrate the Microsoft SQL Server database schemas and data, whilst
Migration providing the ability to specifiy the table geometry during the migration
process.
Utility A representative from the Microsoft Consulting Services will be happy to

run the APS Migration Utility for you please connect with your
Microsoft representative.
Table As we learnt within the previous sections, SQL Server Parallel Data
Warehouse (PDW) is a Massively Parallel Processing (MPP) appliance
Geometry which follows the Shared Nothing Architecture. This means that we need
to distribute or spread the data across the compute nodes in order to
benefit from the storage and query processing of the MPP architecture.
SQL Server PDW provides two options to define how the data can be
partitioned, these are distributed and replicated tables.
Distributed Tables
A distributed table is a table in which all rows have been spread across
the SQL Server PDW Appliance compute nodes based upon a row hash
function. Each row of the table is placed on a single distribution,
assigned by a deterministic hash algorithm taking as input the value
contained within the defined distribution column.
The following diagram depicts how a row would typically be stored

within a distributed table.

Each SQL Server PDW compute node has eight distributions and each
distribution is stored on its own set of disks, therefore ring-fencing the
I/O resources.
Distributed tables are what gives SQL Server PDW the ability to scale out
the processing of a query across multiple compute nodes.
Each distributed table has one column which is designated as the

distribution column. This is the column that SQL Server PDW uses to
assign a distributed table row to a distribution.
When selecting a distribution column there are performance

considerations, one of which is data skew. Skew occurs when the rows of
a distributed table are not spread uniformly across each distribution.
When a query relies on a distributed table which is skewed, even if a
smaller distribution completes quickly, you will still need to wait for the
queries to finish on the larger distributions. Therefore a parallel query
performs as slow as the slowest distribution, and so it is essential to
avoid data skew. Where data skew is unavoidable, PDW can endure 20 -
30% skew between the distributions with minimal impact to the queries.

Replicated Tables
A replicated table is a table in which a complete duplicate of all data is
stored on each of the SQL Server PDW compute nodes. Replicating a
table onto each compute node provides an additional optimization by
removing the need to shuffle data between distributions before
performing a join operation. This can be a huge performance boost
when joining lots of smaller tables to a much larger tables, as would be
the case within a type dimension data model, i.e. distribute the fact table
on a high cardinality column, whilst replicating the dimensional tables.
The above diagram depicts how a row would be stored within a

replicated table. A replicated table is striped across all of the disks
assigned to each of the distributions within a compute node.
Because you are duplicating all data for each replicated table on each
compute node, you will require extra storage, equivalent to the size of a
single table multiplied by the number of compute nodes, therefore a
table containing 100MB of data on a PDW appliance with 10 compute
nodes will require 1GB of storage.
When loading new data, or updating existing data you will require far
more resources to complete the task, than if it were to be executed
against a distributed table. This is due to the fact that each operation
will need to be executed on each compute node. Therefore it is essential
that you take into account the extra overhead when performing ETL/ELT
style operations against a replicated table.

The most frequently asked question about replicated tables would be
what is the maximum size we should consider for a replicated table?
We recommend that you keep the use of replicated tables to a small set
of data, generally less than 5GB in size. However there are scenarios
when you may want to consider much larger replicate tables. In which
case we would recommend you follow one of these approaches to
reduce the overall batch resource requirements.
Maintain 2 versions of the same table, one distributed in which to

perform all of the ETL/ELT operations against, and the second
replicated for use by the BI presentation layer. Once all of the
ETL/ELT operations have completed against the distributed version
of the table, you would then execute a single CREATE TABLE AS
SELECT (CTAS) statement in which to rebuild a new version of the
final replicate table.
Maintain 2 tables, both replicate tables. One base table persists

all data, minus the current weeks set of changes. The second
delta table persists the current weeks set of changes. A view is
created over the base and the delta replicate tables which
resolves the two sets of data into the final representation. A weekly
schedule is required, executing a CTAS statement based upon the
view to rebuild a new version of the base table, once complete
the delta table can be truncated.
The Design Rules

A small number of database design choices can have a big impact on
improving query performance. The following design rules will provide
you with an outline to what information you should look for when
making design decisions.
The key to achieving optimal performance using SQL Server Parallel

Data Warehouse (PDW) is to make full use of the Massively Parallel
Processing (MPP) architecture. You can achieve this by following two
golden rules which are:
DISTRIBUTION FIRST
What we mean by Distribution First is that you should always begin
with designing your data model to be distributed, as if you do not, you
will not be able to realize the benefits of the Massive Parallel Processing
(MPP) architecture. Therefore always begin with a distributed table.
Replicated tables should be viewed as an optimization technique, much

in the same way you would view indexing or partitioning.

When designing your data model consider the following points:
A single subject area within a data model should have a common

distribution column which is also part of the join condition
De-normalization with a Clustered ColumnStore Index as a way to

overcome poor distribution column selection options.
MINIMIZE ALL MOVEMENT OF DATA

Data Movement comes in many forms such as:
From physical disk to memory
Redistribution of data between distributions as part of a SHUFFLE

MOVE operation
Duplication of data between compute nodes as part of a

BROADCAST MOVE operation
We should attempt to minimize as much movement as possible to

achieve the best possible performance. This can be achieved by better
selection of distribution columns and optimization techniques.
Some data movement may be necessary to satisfy a query. It is worth

mentioning that the PDW engine and Data Movement Service (DMS)
does an excellent job of minimizing data movement operations using
predicate pushdown, local/global aggregation and table statistics. So it
is very important to keep PDW statistics up-to-date.
Distribution Column Selection Rules

Selecting a good distribution column is an important aspect to
maximizing the benefits of the Massive Parallel Processing (MPP)
architecture. This is possibly the most important design choice you will
make in SQL Server Parallel Data Warehouse (PDW).
The principal criteria you must consider when selecting a distribution

column for a table are the following:
Access
Distribution
Volatility
The ideal distribution key is one that meets all three criteria. The reality
is that you are often faced with trading one off from the other.

JOIN ACCESS - Select a distribution column that is also used within
query join conditions
Consider a distribution column that is also one of the join conditions
between two distributed tables within the queries being executed. This
will improve query performance by removing the need to redistribute
data via a SHUFFLE MOVE operation, making the query join execution
distribution local.
AGGREGATE ACCESS Select a distribution column that is also

aggregation compatible
Consider a distribution column that is also commonly used within the
GROUP BY clause within the queries being executed. This will improve
query performance by removing the need to perform a two-step
aggregation and redistribute data via a SHUFFLE MOVE operation,
making the query GROUP BY operation execute distribution local.
DISTINCT ACCESS Select a distribution column that is frequently used

in COUNT DISTINCTs
Consider a distribution column that is commonly used within a COUNT
DISTINCT function. This will improve query performance by removing
the need to redistribute data via a SHUFFLE MOVE operation, making
the query COUNT DISTINCT function execute distribution local.
DATA DISTRIBUTION Select a distribution column that provides an

even data distribution
Consider a distribution column that can provide an even number of rows
per distribution, therefore balancing the resource requirements. To
achieve this look for a distribution column that provides a large number
of distinct values, i.e. at least 10 times the number of table distributions.
The data distribution of a table is said to be skewed if one or more

distributions contain significantly more rows than others. Data skew is
the enemy of all Massively Parallel Processing (MPP) systems as it puts
the system into an unbalanced state which requires more overall
resources within one or more distributions. When selecting a distribution
column ensure that no single value represents a large percentage of
rows (including NULLs). Unfortunately it is highly unlikely to be able to
eliminate all data skew, but when selecting a distribution column, always
remember, the greater the data skew, the greater the performance
impact.
VOLATILITY Select a distribution column that rarely, if ever changes

value
PDW will not allow you to change the value of a distribution column for
a given row by using an UPDATE statement. This is due to the fact that
changing the value of a distribution column for a given row will almost
certainly mean the row will be moved to a different distribution.

If you are required to change the value of a distribution column then
you will need to first delete the existing row and then insert the new
row. Alternatively you could use a CREATE TABLE AS SELECT (CTAS)
statement to rebuild the table or partition.
It is therefore recommended that you select a distribution column which

rarely requires the value to be modified.
Optimization SQL Server Parallel Data Warehouse (PDW) is a Massively Parallel

Processing (MPP) appliance which provides you with large amounts of
within PDW resources however these are still finite and overall query and
concurrency performance will benefit from optimization of your data
model using one or more of the following options.
Replicated Table
A replicated table is a table in which a complete duplicate of all data is
stored on each of the SQL Server PDW compute nodes. Replicating a
table can provide the ability for it to become distribution compatible
when joined to another table. This removes the need to perform data
movement via a SHUFFLE MOVE operation at the expense of data
storage and load performance.
The ideal candidate for replicating a table is one that is small in size, i.e.
less than 5GB, changes infrequently and has been proven to be
distribution incompatible. Before selecting to replicate a table, ensure
that you have exhausted all options for keeping the table distributed.
The dimensional model above depicts one fact table and four dimension
tables. Within this example the customer dimension is significantly larger

in size than all other dimensions and based upon this would benefit
remaining distributed. The Customer ID column provides good data
distribution (minimal data skew), is part of the join condition between
the fact table and the customer dimension creating a distribution
compatible join, and is a regular candidate to be within the GROUP BY
clause during aggregation. Therefore selecting Customer ID as the
distribution column would be a good candidate. All other dimensions
are small in size and to remove any need to redistribute the fact table
when performing a join, we have made them replicated. Therefore
eliminating all data movement when joining the fact table to any
dimension table(s).
Clustered Indexes
A Clustered Index physically orders the pages of the data in the table. If
the table is distributed, then the physical ordering of the data pages is
applied to each of the distributions individually. If the table is replicated,
then the physical ordering of the pages is applied to each of replicate
tables on each of the compute nodes.
All data pages within a clustered index table are linked to the next and
previous data pages in a doubly linked list to provide ordered scanning.
In other words, the records in the physical structure are sorted according
to the elds that correspond to the columns used in the index.
PDW clustered index tables are always defined as non-unique, therefore

each record will contain a 4-byte value (commonly known as an
uniquifier) added to each value in the index where duplicate values
exist.
You can only have one clustered index on a table, because the table
cannot be ordered in more than one direction.
The following diagram shows what a Clustered Index table might look
like for a table containing city names:

When defining a clustered index on a distributed table within PDW it is
not necessary to include the distribution column within the cluster key.
The distribution column can be defined to optimize the joins while the
clustered index can be defined based upon the most frequently used
columns as predicates or range operations.
Clustered indexes within PDW provides you with a way to optimize a

number of standard OLAP type operations, including the following:
PREDICATE QUERIES
Data within the table is clustered on value and an index tree constructed
for direct access to these data pages, therefore clustered indexes will
provide the minimal amount of I/O required in which to satisfy the
needs of a predicate query. Consider using a clustered index on
column(s) commonly used as a valued predicate.
RANGE QUERIES
All data within the table is clustered and ordered on value, which
provides an efficient method to retrieve data based upon a range query.
Therefore consider using a clustered index on column(s) commonly used
within a range predicate, for example where a given date is between two
dates.

AGGREGATE QUERIES
All data within the table is clustered and ordered on value, which
removes the need for a sort to be performed as part of an aggregation
or a COUNT DISTINCT. Therefore consider using a clustered index on
column(s) commonly contained within the GROUP BY clause or COUNT
DISTINCT function.
Clustered Indexes should be selected over Non-Clustered Indexes when

queries commonly return large result sets.
Non-Clustered Indexes
Non-Clustered Indexes are fully independent of the underlying table and
up to 999 can be applied to both heap and clustered index tables. Unlike
Clustered Indexes, a Non-Clustered Index is completely separate from
the data, and on the index leaf page, there are pointers to the data
pages. The pointer from the index to a data row is known as a row
locator. The row locator is structured differently depending on whether
the underlying table is a heap table or a clustered index table.
The row locator for a heap table will contain a pointer to the actual data
page, plus the row number within the data page. If that row is updated
and a data page split is required (if the updated row requires additional
space), a forwarding pointer is left to point to the new data page and
row. Over time and frequent updates can cause poor performance of the
Non-Clustered Indexes, requiring maintenance via a rebuild of the
underlying table and indexes.
The row locator for a clustered index table is different as it will contain
the cluster key from the cluster index. To find a value, you first traverse
the non-clustered Index, returning the cluster key, which you then use to
traverse the clustered index to reach the actual data. The overhead to
performing this is minimal as long as you keep the clustering key
optimal. While scanning two indexes is more work than just having a
pointer to the physical data page location, overall it can be considered
better because minimal reorganization is required for any modification
of the values in the table. This benet is only true if the cluster key rarely,
or never, changes.
The same reasons for selecting a clustered index are also true for a non-
clustered index with the additional rules that they should only be used
to optimize queries which return small result sets or when they could be
utilized for covering indexes, reducing the base table IO requirements
(However with the advent of Clustered ColumnStore Indexes this is less
relevant).
NOTE: Non-Clustered Indexes can contribute to fragmentation and

random IO, which goes against the sequential IO requirements of data
warehousing. It is therefore recommended that you always begin with
no Non-Clustered Indexes and add them only if its really necessary.

Clustered ColumnStore Indexes
A Clustered ColumnStore Index uses a technology called xVelocity for
the storage, retrieval and management of data within a columnar data
format, which is known as the columnstore. Data is compressed, stored,
and managed as a collection of partial columns, which we call column
segments.
Some of the clustered columnstore index data is stored temporarily

within a rowstore table, known as a deltastore, until it is compressed and
moved into the columnstore. The clustered columnstore index operates
on both the columnstore and the deltastore, when returning the query
results.
KEY CHARACTERISTICS
A clustered columnstore index in SQL Server Parallel Data Warehouse
(PDW) has the following characteristics:
Available in SQL Server PDW and above.
Includes all columns in the table and is the method for storing the
entire table.
Fully updateable.
Can be partitioned.
It is the only index on the table, therefore cannot be combined with

any other indexes.
Uses columnstore compression which is not configurable.
Does not physically store columns in a sorted order. Instead, it

stores data to improve compression and performance. Pre-sorting
of data can be achieved by creating the Clustered ColumnStore
Index table from a Clustered Index table
SQL Server PDW moves the data to the correct location

(distribution, compute node) before adding it to the clustered
columnstore index.
For a distributed table, there is one clustered columnstore index for

every partition of every distribution.
For a replicated table, there is one clustered columnstore index for

every partition of the replicated table on every compute node.

BENEFITS
SQL Server Parallel Data Warehouse (PDW) takes advantage of the
column based data layout to significantly improve compression rates
and query execution time.
Columns often have similar data, which results in high compression

rates.
Higher compression rates improve query performance by requiring

less total I/O resources and using a smaller in-memory footprint.
A smaller in-memory footprint allows for SQL Server PDW to

perform more query and data operations in-memory.
Queries often select only a few columns from a table, requiring less
total I/O resources.
Columnstore allows for more advanced query execution to be

performed, by processing the columns within batches, which
reduces CPU usage.
KEY TERMS
The following are key terms and concepts that you will need to know in
order to better understand how to use clustered columnstore indexes.
Row Group is a group of rows that are compressed into

columnstore format at the same time. Each column in the row
group is compressed and stored separately on the physical media.
Column Segment a column segment is the basic storage unit for

a columnstore index. It is a group of column values that are
compressed and physically stored together on the physical media.
ColumnStore a columnstore is data that is logically organized as

a table with rows and columns, physically stored in a columnar data
format. The columns are divided into segments and stored as
compressed column segments.
RowStore a rowstore is data that is organized as rows and

columns, and then physically stored in a row based data format.
DeltaStore a deltastore is a rowstore table that holds rows until

the quantity is large enough to be moved into the columnstore.
When you perform a bulk load, most of the rows will go directly to
the columnstore without passing through the deltastore. Some
rows at the end of the bulk load might be too few in number to
meet the minimum size of a rowgroup. When this happens, the
final rows go to the deltastore instead of the columnstore.

Tuple Mover a background process which automatically
compresses CLOSED Row Groups from the DeltaStore into
Column Segments and storing them in the ColumnStore.
Partitioning
Partitioning allows you to physically divide up tables and indexes
horizontally, so that a groups of data are mapped into individual
partitions. Even though the table has been physically divided, the
partitioned table is treated as a single logical entity when queries or
updates are performed on the data.
BENEFITS
Partitioning large tables within SQL Server Parallel Data Warehouse
(PDW) can have the following manageability and performance benefits.
SQL Server PDW automatically manages the placement of data in

the proper partitions.
A partitioned table and its indexes appear as a normal database

table with indexes, even though the table might have numerous
partitions.
Partitioned tables support easier and faster data loading, aging,

and archiving. Using the sliding window approach and partition
switching.
Application queries that are properly filtered on the partition

column can perform better by making use of partition elimination
and parallelism.

You can perform maintenance operation on partitions, efficiently
targeting a subset of data to defragment a clustered index or
rebuilding a clustered columnstore index.
SQL Server Parallel Data Warehouse (PDW) simplifies the creation of

partitioned tables and indexes as you are no longer need to create a
partition scheme and function. SQL Server PDW will automatically
generate these and ensure that data is spread across the physical disks
efficiently.
For distributed tables, table partitions determine how rows are grouped
and physically stored within each distribution. This means that data is
first moved to the correct distribution before determining which
partition the row will be physically stored.
For clustered columnstore indexes, every partition will contain their own
columnstore and deltastore. To achieve the best possible performance
and maximise compression, it is recommended that you ensure each
partition for each distribution is sized so that it contains more than 1
million rows.
Within SQL Server you are able to create a table with a constraint on the
partitioning column and then switch that table into a partition. Since
PDW does not support constraints, we do not support this type of
partition switching method, but rather would require the source table to
be partitioned with matching ranges.
Statistics Collection
SQL Server Parallel Data Warehouse (PDW) uses a cost based query
optimizer and statistics to generate query execution plans to improve
query performance. Up-to-date statistics ensures the most accurate
estimates when calculating the cost of data movement and query
operations.
SQL Server PDW stores two sets of statistics at different levels within the
appliance. One set exists at the control node and the other set exists on
each of the compute nodes.
COMPUTE NODE STATISTICS

SQL Server PDW stores statistics on each of the compute node, and uses
them to improve query performance for the queries which execute on
the compute nodes.
Statistics are objects that contain statistical information about the

distribution of values in one or more columns of a table. The cost based
query optimizer uses these statistics to estimate the cardinality, or
number of rows, in the query result. These cardinality estimates enable
the query optimizer to create a high-quality query execution plan.

The query optimizer could use cardinality estimates to decide on
whether to select the index seek operator instead of the more resource-
intensive index scan operator, improving query performance.
Each compute node has the AUTO_CREATE_STATISTICS set to ON, which

causes the query optimizer to create statistics based upon a single
column that is referenced within the WHERE or ON clause of a query.
Multi-Column statistics are not automatically created.
CONTROL NODE STATISTICS

SQL Server PDW stores statistics on the control node, and uses them to
minimize the data movement in the distributed query execution plan.
When you create statistics on a distributed table, SQL Server PDW

creates a statistics object for each distribution table on each of the
compute nodes, and one statistics object on the control node. The
control node statistics object contains the result from aggregating the
statistics object created for each of the distributions.
When you create statistics on a replicated table, SQL Server PDW creates
a statistics object for each compute node, but since each compute node
will contain the same statistics, the control node will only copy one
statistics object from one compute node.
When you create statistics on an external table, SQL Server PDW will first
import the required data into PDW so that it can then compute the
statistics. The results are stored on the control node.
Statistics stored on the control node are not made available to client
applications. The DMVs report only statistics on the compute nodes and
so do not report statistics on the control node.
Collection of statistics can be performed based upon a default sample,

custom sample percentage, filter selection or a full table scan. Increasing
the amount of data sampled improves the accuracy of the cardinality
estimates at the expense of the amount of time required in which to
calculate the statistics.
Collection of Multi-Column statistics on all join columns within a

single join condition and all aggregate columns within a GROUP BY
clause will provide additional information for the optimizer to generate a
better execution plan, such a density vectors.
Always ensure that statistics do not go stale, by updating the statistics

automatically as part of the transformation schedule or a separate
maintenance schedule, depending on the volatility of the column or
columns.
As a minimum statistics should be created for each of the following:

Distribution Column
Partition Column
Clustered Index Keys
Non-Clustered Index Keys
Join Columns
Aggregate Columns
Commonly used predicate columns
This section discusses migration issues that can be encountered when

Converting converting schema objects from SQL server to PDW and how they can
be overcome.
Database
Schema
Objects
Type Migrating SQL Server Data Types is relatively easy as most data types
are the same between SQL Server and PDW.
Mapping Equivalent Data Types

The following data types are the same between SQL Server and PDW.
binary, bit, char, date, datetime, datetime2,

datetimeoffset, decimal, float, int, bigint, money,
nchar, nvarchar, real, smalldatetime, smallint,
smallmoney, time, tinyint, varbinary, varchar
Data Types Not Supported
cursor, image, ntext, numeric, rowversion,

hierarchyid, table, text, timestamp, xml,
uniqueidentifier, Spatial Types, CLR data types
Data Type Migration Considerations and Mitigations

The following data types have differences or are not supported.
NUMERIC

The Numeric data type is the same as the Decimal data type in SQL
server. In PDW the name numeric has been deprecated and PDW uses
decimal instead. The way to convert from numeric to decimal is to
change the definition name from numeric to decimal as the types are
functionally equivalent.
CREATE TABLE dbo.Purchases

(
PurchaseOrderID int NOT NULL,
Value Numeric(9)
);
Translates to:
CREATE TABLE dbo.Purchases

(
PurchaseOrderID int NOT NULL,
Value Decimal(9)
);
DATE AND TIME FORMATS

The Date and Time data types are treated the same way in SQL Server
and PDW. So data in PDW or SQL Server databases can be queried the
same way and follows the same rules as far as both storage and syntax.
However its possible that issues may arise when loading data. In many
countries the date and time formats are different from the default of
ymd. In those cases the format needs to be handled in order to avoid
loading errors via DWloader. DWloader is one of the ways to load data
into PDW from text files. Text files however do not contain any type
information and so its possible that the format may be miss interpreted
and cause errors forcing rows to be skipped (rejected) and logged
during data loads. Dwloader.exe has numerous parameters in order to
handle the type conversion into the appropriate format.
Example:
dwloader.exe S MyPDW-SQLCTL01 -U mylogin -P 123jkl

i file.txt -t "|" -r \r\n D "dmy"
The date format in this case assumes that all the columns have the date
values ordered as day (d), month (m), year (y). Its also possible that
formats differ between columns. In that case a format file will have to
be used that specifies the input format on a per column basis. The
format file can be specified with the dt parameter.
dwloader.exe S MyPDW-SQLCTL01 -U mylogin -P 123jkl

i file.txt -t "|" -r \r\n dt inputformatfile.txt

The contents of the inputformatfile.txt would look like this. Each line has
the name of the column and the format of the destination datetime data
type.
LastReceiptDate=ymd
ModifiedDate=mdy
When importing text data its also possible that due to errors in the
source data the date format for a specific column varies from row to row
and this can either lead to errors or loading incorrect data. There may
also be corruption errors in the input data that need to be detected. In
such cases dwloader.exe offers some options to detect and log the rows
that caused a load error. This would then allow bayou to remediate the
data and reload.
In these cases the following dwloader parameters are useful. The

complete list of dwloader parameters is available in the product books
online.
-rt { value | percentage }

Specifies whether the -reject_value in the -rv reject_value option
is a literal number of rows (value) or a rate of failure
(percentage). The default is value. The percentage option is a
real-time calculation that occurs at intervals according to the -rs
option. For example, if the Loader attempts to load 100 rows
and 25 fail and 75 succeed, then the failure rate is 25%.
-rv reject_value
Specifies the number or percentage of row rejections to allow
before halting the load. The -rt option determines if reject_value
refers to the number of rows or the percentage of rows. The
default reject_value is 0. When used with -rt value, the loader
stops the load when the rejected row count exceeds reject_value.
When use with -rt percentage, the loader computes the
percentage at intervals (-rs option). Therefore, the percentage of
failed rows can exceed reject_value.
-rs reject_sample_size
Used with the -rt percentage option to specify the incremental
percentage checks. For example, if reject_sample_size is 1000, the
Loader will calculate the percentage of failed rows after it has
attempted to load 1000 rows. It recalculates the percentage of
failed rows after it attempts to load each additional 1000 rows.
-R load_failure_file_name
If there are load failures, dwloader stores the row that failed to
load and the failure description the failure information in a file
named load_failure_file_name. If this file already exists, dwloader

will overwrite the existing file. load_failure_file_name is created
when the first failure occurs. If all rows load successfully,
load_failure_file_name is not created.
-e character_encoding
Specifies a character-encoding type for the data to be loaded
from the data file. Options are ASCII (default), UTF8, UTF16, or
UTF16BE, where UTF16 is little endian and UTF16BE is big endian.
These options are case insensitive.
-m
Use multi-transaction mode for the second phase of loading;
when loading data from the staging table into a distributed
table.
With m, SQL Server PDW performs and commits loads in

parallel. This performs much faster than the default loading
mode, but is not transaction-safe.
Without m, SQL Server PDW performs and commits loads

serially across the distributions within each Compute node, and
concurrently across the Compute nodes. This method is slower
than multi-transaction mode, but is transaction-safe.
-m is optional for append, reload, and upsert.
-m is required for fastappend.
-m cannot be used with replicated tables.
-m applies only to the second loading phase. It does not apply to

the first loading phase; loading data into the staging table.
There is no rollback with multi-transaction mode, which means

that recovery from a failed or aborted load must be handled by
your own load process.
We recommend using m only when loading into an empty

table, so that you can recover without data loss. To recover from
a load failure: drop the destination table, resolve the load issue,
re-create the destination table, and run the load again.
TIMESTAMP
Timestamp in PDW is treated differently to SQL Server. In SQL server the
current practice is to use the rowversion keyword instead of the
timestamp keyword however the two commands are functionally
equivalent in SQL Server. In PDW there is no direct functional equivalent
to the rowversion that auto updates if the row is changed. In PDW
partially similar functionality can be achieved by using the datetime2

data type and CURRENT_TIMESTAMP. Note that in PDW the column
does not auto update if the row is changed like it does in SQL Server.
In SQL Server rowversion and timestamp columns are used to provide

application controlled row change detection and optimistic locking for
mainly OLTP databases. In PDW the same is not true since the platform
is designed to support data warehouse requirements rather than
thousands of singleton insert and updates.
The issue can sometimes arise when data needs to be migrated to the
data warehouse and a column is needed to contain the source
rowversion data. The binary data type can provide a solution here.
Non nullable rowversion / timestamp data type is equivalent to

BINARY(8) data type.
Nullable rowversion / timestamp data type is equivalent to

VARBINARY(8) data type.
If this functionality is required then either the application will need to

track and make the relevant updates in application logic or this
functionality could be undertaken in SQL Server and the resulting data
loaded into PDW.
For example:
CREATE TABLE Table1 (Key1 INT, VersionCol

ROWVERSION);
Can be converted to the following in PDW:
CREATE TABLE Table1 (Key1 INT, VersionCol BINARY(8));
There are also cases where a timestamp is needed in order to track a

certain action or event or even provide for some rudimentary application
based locking behavior. Although it is not a best practice to carry out
OLTP style workloads within a data warehouse there may be no other
easy option to overcome a specific issue. A common use of this is to be
able to easily record the batches of data loaded into the data warehouse
or a specific multistep ELT process.
For example:
CREATE TABLE ProcessRunning

(
BatchID INTEGER,
DescriptionVal VARCHAR(30),
col2 DATETIME2
);
DECLARE @currTS DATETIME2

SET @currTS = CURRENT_TIMESTAMP
INSERT INTO ProcessRunning VALUES (1, 'BIG ELT Xform
process', @currTS);
-- The BatchID could then be used to identify added
rows in a given batch.
MIGRATING TEXT, BLOBS AND VARCHAR(MAX), NVARCHAR(MAX),

VARBINARY(MAX)
The following data types are not supported in PDW:
TEXT
VARCHAR(MAX)
NVARCHAR(MAX)
VARBINARY(MAX)
XML
When dealing with these data types there are several options that need
to be evaluated.
1. Leave the large BLOB data in the source operational systems.
2. Break up BLOB data into logical fields and then migrate. In many
cases VARCHAR(MAX) columns are being used to store data that
could be broken down further into different columns and different
data types. While this is not always possible a simple investigation
of the actual data in the VARCHAR(MAX) fields can yield a lot of
information and provide more choices. When following this
approach you also need to consider the row size limitations.
3. Determine the actual size of the largest amount data in the

VARCHAR(MAX) field. In most cases this is a small enough value
that the data could be translated to a smaller data type.
For example:
-- Displays the max length of the largest string.

SELECT MAX(DATALENGTH(DescriptionVal)) FROM tbl1
If the size of the data is still too big to fit in the row and including it
would exceed the 8KB per page limit then another potential option is to
have the large text column in a separate table and related to the main
row by an ID.

Migrating or Working with Spatial Data Types
In the current release PDW does not support directly performing
calculations with spatial data types in the same way as is possible with
SQL Server. Having said this it is possible to store spatial data in PDW
so that it can be retrieved and processed in a seamless fashion in order
perform further analysis by utilizing a distributed database architecture.
In this case PDW would be the master system hosting the source raw
data. SQL server hosts could then extract and post process this data to
perform other calculations. Fast Infiniband interconnects provide
sufficient throughput to transfer large volumes of data at a high speed
in order to both reduce the burden on the central enterprise data
warehouse as well as harness native functionality that is available in SQL
server today.
Base raw data can be migrated from SQL Server and stored in existing
PDW data types with no loss. The data can be retrieved via normal T-
SQL queries and the spatial processing can be done via SQL server.
Importing Data - Spatial Data can be converted into data which can then
be stored within a PDW compatible data type such as an NVARCHAR()
or VARCHAR(). Importing Spatial Types and converting them to
VARCHAR can be accomplished via SSIS, CLR or custom data conversion
application/script.
Retrieving Data - SQL Server can connect to PDW via a linked server
connection, querying data directly or retrieving data via SSIS. Both
connection types operate over the fast Infiniband network to support
fast data transfers.
Linked servers can be queried within SQL server in much the same
fashion as tables. A view would then be created to convert back to a
spatial type to support geometric calculations within SQL Server. Overall
this solution would be cost effective to implement since the heavy duty
data retrieval and aggregations work is done by PDW in a scalable
fashion. Geospatial calculations and any post processing can be
performed on less expensive SMP servers running SQL Server Standard
Edition.

This is an example of working with spatial data.
Syntax There are syntax differences to be aware of between SQL Server and
PDW. In most cases the extra complexity needed in the SQL Server
Differences syntax has been removed as its not applicable for PDW. This reduced
syntax leveraging the appliance model is both easier to learn as well as
less complex to support in the long term.
Create Database Statement

The create database statement syntax is significantly simpler than the
syntax needed for SQL Server.
CREATE DATABASE database_name

WITH (
[ AUTOGROW = ON | OFF , ]
REPLICATED_SIZE = replicated_size [ GB ] ,
DISTRIBUTED_SIZE = distributed_size [ GB ] ,
LOG_SIZE = log_size [ GB ] )
[;]
The appliance automatically places the database files on the

appropriately for best performance and therefore the file specification
and other file related parameters are not needed. This paper has
already covered concepts relating to distributed tables vs replicated
tables in previous sections.

REPLICATED_SIZE
The amount of data in GB allocated for replicated tables. This is
the amount of data allocated on a per compute node basis.
DISTRIBUTED_SIZE
The amount of data in GB allocated for distributed tables. This is
the amount of data allocated over the entire appliance. Meaning
that in a on a per compute node basis the disk space allocated
equals the DISTRIBUTED_SIZE divided by the number of compute
nodes in the appliance.
LOG_SIZE
The amount of data in GB allocated for the transaction log. This
is the amount of data allocated over the entire appliance.
Meaning that in a on a per compute node basis the disk space
allocated equals the LOG_SIZE divided by the number of
compute nodes in the appliance.
AUTOGROW
Governs if the database size is fixed or if the appliance will
automatically increase the disk allocation for the database as
needed until the amount of physical disk space in the appliance
is consumed. It is recommended that the database be created
big enough before loading data in order to minimize auto grow
events that can impact data loading performance.
MIGRATING A SQL SERVER DATABASE AND SIZING FOR PDW

When migrating databases one must calculate the amount of disk space
allocated for replicated vs distributed tables. These general rules
provide guidance for when to make a table replicated or distributed.
When to set tables as replicated.
Dimension and reference tables are typically replicated.
Tables with less than 5GB of heap data are typically replicated.
Tables that have no identifiable distribution key that also match

the other criteria.
When to set tables as distributed.
Fact tables should be distributed.
Tables with greater than 5 GB of heap data (excluding indexes)

is considered large enough to typically be set as distributed.

Tables that have columns with reasonably high cardinality and
where these columns are ideally also present in aggregation
clauses in queries submitted.
Understanding the business and data will also highlight

distributed tables.
Once the table sizes are known and categorized then its a simple matter
of adding a percentage for growth on a year to year basis. It is also
important to allow sufficient space to be able to make a copy of the
largest distributed table should the need arise. If partitioning is used on
the fact table(s) since its so large then sufficient space for at least a few
extra partitions should be allocated. These simple guidelines together
with the categorization of tables between distributed vs. replicated will
provide sufficient insight into sizing the database.
Create Table Statement

The main differences in syntax relating to the create table statement is
the extra syntax related to nominating a table as either replicated or
distributed or if the table is stored as a clustered column store.
For example:
CREATE TABLE myTable

(
id INT NOT NULL
, lastName VARCHAR(20)
, zipCode VARCHAR(6)
)
WITH
(
DISTRIBUTION = HASH (id) -- <- Nominates the
distribution column
, CLUSTERED COLUMNSTORE INDEX -- <- Specifies
column store.
);
Replicated tables are defined with this syntax:
WITH
(
DISTRIBUTION = REPLICATE -- <- Nominates the
TABLE as replicated
);

Create Table Differences
PRIMARY KEY and UNIQUE syntax is not supported.
Data compression is always page compression unless the table is

created with a column store structure.
FILESTREAM storage is not supported.
Cannot specify that a table is to be created in a specific filegroup

by naming the filegroup. Partitions can be used to spread and
segment data. This reduces the effort required by the DBA.
Referential FOREIGN KEY constraints cannot be defined. However

default constraints are supported. In data warehouse designs
referential integrity constraints impose a very significant
performance penalty when loading data and updating data as the
constraints have to be validated. Current practice with enterprise
data warehouse systems is to avoid the use of referential
constraints in favor of higher performance.
Cascade update and delete operations are not supported.
Specifying Partitions
Partitions must be specified as part of the create table statement. The
partition scheme and partition function syntax that is required for SQL
Server is not required for PDW. The partition functions and schemes are
setup automatically by PDW.
For example:
The following is SQL Server Syntax:
CREATE PARTITION FUNCTION myRangePF1 (int)

AS RANGE LEFT FOR VALUES (1, 100, 1000)
;
GO
CREATE PARTITION SCHEME myRangePS1

AS PARTITION myRangePF1
TO (test1fg, test2fg, test3fg, test4fg)
;
GO

(
ID INTEGER,
Col1 VARCHAR(10),
Col2 VARCHAR(10)
)
ON myRangePS1 (ID)
;

Translates to the following PDW Syntax (assuming a distributed table):

(
ID INTEGER,
Col1 VARCHAR(10),
Col2 VARCHAR(10)
)
WITH
(
DISTRIBUTION = HASH(ID)
, PARTITION (ID RANGE LEFT FOR VALUES (1, 100,
1000))
);
The $PARTITION function is not supported in queries however partition

elimination is still supported and performed via the use of WHERE clause
predicates.
For example:
The following is SQL Server Syntax:
SELECT COUNT(*) FROM myTable

WHERE col1 = "xx"
AND $PARTITION.myRangePF1(ID) = 2
;
Translates to the following PDW Syntax:
SELECT COUNT(*) FROM myTable

WHERE col1 = "xx"
AND ID > 100 AND ID <= 1000 -- Specifies
partition 2
;
Migrating Triggers
PDW does not support Triggers functionality in the current release.
Triggers are mainly used in OLTP solutions and therefore have very
limited application in data warehouse solutions. Triggers in OLTP
solutions are used for implementing cascading update functionality or
other logic based on changed rows. In data warehouse solutions data
changes can be implemented via operational data stores (ODS) or in the
ETL/ELT processes that cleanse and layout data for maximum retrieval
query and processing performance.
It is recommended that functionality requiring triggers is implemented

in ODS systems and not in the enterprise data warehouse. A hub and
spoke architecture using both PDW and SQL Server would be an ideal
solution to such issues. Intermediate work can be performed on SMP
SQL servers and the heavy duty queries and aggregations can be
executed against PDW.

Schema Ownership
Schemas operate in a similar fashion in PDW as they do in SQL Server. In
SQL Server when creating a schema within the same transaction you can
also create tables, views, and set GRANT, DENY or REVOKE permissions
on those objects. This is not supported in PDW.
SQL Server CREATE SCHEMA:
USE AdventureWorks2012;
GO
CREATE SCHEMA Sprockets AUTHORIZATION mwinter
CREATE TABLE NineProngs
(
source int
, cost int
, partnumber int
)
GRANT SELECT ON SCHEMA::Sprockets TO julius
DENY SELECT ON SCHEMA::Sprockets TO danny;
GO
Equivalent PDW Syntax:
USE AdventureWorks2012;
CREATE SCHEMA Sprockets AUTHORIZATION mwinter
GO
CREATE TABLE Sprockets.NineProngs
(
source int
, cost int
, partnumber int
)
WITH (DISTRIBUTION = HASH(source))
GRANT SELECT ON SCHEMA::Sprockets TO julius
DENY SELECT ON SCHEMA::Sprockets TO danny
GO
Views
Views operate in a similar fashion in PDW as they do in SQL Server. In
PDW the view definitions are stored in the control node and expanded
during query compilation. As such schema binding and materialized
views are not supported.
Unsupported View Creation Options:
WITH CHECK OPTION, ENCRYPTION, SCHEMABINDING,

VIEW_METADATA
Migrating SELECT Statements

The select statement between SQL Server and PDW is very similar. Minor
additions exist in order to harness the added performance that PDW can
provide when selecting and transforming large amounts of data. PDW
introduces a new statement called Create Table as Select (CTAS). This

replaces the SQL Server SELECT INTO statement and provides additional
functionality, such as the ability to specify the table geometry
(Distributed, Replicate), indexing (Heap, Clustered Index, Clustered
ColumnStore Index) and partitioning options. This operation is also
minimal logged by default.
Changing SELECT INTO syntax to CTAS Statements will leverage the

added performance that PDW provides. The CTAS statement is covered
in other sections of this document so this example focuses on the syntax
differences.
For example, the SQL Server SELECT INTO Statement:
SELECT
c.ID
, c.FirstName
, c.LastName
, e.JobTitle
, a.AddressLine1
, a.City
, a.PostalCode
INTO dbo.AddressesList
FROM Person.Person AS c
INNER JOIN HumanResources.Employee AS e
ON e.BusinessEntityID = c.BusinessEntityID
INNER JOIN Person.BusinessEntityAddress AS bea
ON e.BusinessEntityID = bea.BusinessEntityID
INNER JOIN Person.Address AS a
ON bea.AddressID = a.AddressID
INNER JOIN Person.StateProvince as sp
ON sp.StateProvinceID = a.StateProvinceID
Translates to the following PDW statement (assuming the table is

distributed on the ID column):
CREATE TABLE dbo.AddressesList

WITH (DISTRIBUTION = HASH (ID))
AS
SELECT
c.ID
, c.FirstName
, c.LastName
, e.JobTitle
, a.AddressLine1
, a.City
, a.PostalCode

Assuming a replicated table, the translation would be:
CREATE TABLE dbo.AddressesList

WITH (DISTRIBUTION = REPLICATE)
AS
SELECT
c.ID
, c.FirstName
, c.LastName
, e.JobTitle
, a.AddressLine1
, a.City
, a.PostalCode
The body of the select statement remains the same within the PDW
syntax translation. All that has changed is the addition of the table
creation portion of the syntax. These relatively minor differences in
syntax can provide orders of magnitude better performance between
SMP and MPP platforms.
Migrating UPDATE Statements

The update statement in PDW is similar to the update statement
functionality in SQL Server in most ways. However there are some
differences that will require modification of update statement syntax.
Update statements cannot contain the TOP clause.
Update statements cannot contain a FROM clause and a sub-query.

For example, the following statement is not allowed:
UPDATE r
SET update_ts = CURRENT_TIMESTAMP
FROM region AS r
WHERE EXISTS (SELECT 1
FROM nation AS n
WHERE n.region_key = r.region_key
AND n.nation_key = 0)
The above query can be rewritten as an implicit join:
UPDATE region
SET update_ts = CURRENT_TIMESTAMP
FROM nation AS n
WHERE n.region_key = region.region_key
AND n.nation_key = 0

Update statements cannot contain explicit joins in the FROM
clause. For example, the following statement is not allowed:
UPDATE c
SET population_ct = population_ct + 1
FROM census AS c
INNER JOIN location AS l
ON c.location_key = l.location_key
WHERE l.country_nm = 'Australia'
AND l.state_nm = 'Victoria'
The above query can be rewritten as an implicit join in the FROM

clause. The following statement is allowed:
UPDATE census
SET population_ct = population_ct + 1
FROM location AS l
WHERE census.location_key = l.location_key
WHERE l.country_nm = 'Australia'
AND l.state_nm = 'Victoria'
Update statements cannot contain more than one join in the FROM
clause. For example, the following statement is not allowed:
UPDATE order_line
SET dispatch_ts = CURRENT_TIMESTAMP
FROM customer AS c
, order AS o
WHERE order_line.order_key = o.order_key
AND order_line.customer_key = c.customer_key
AND o.order_no = 'MS12345678-90'
AND c.customer_nm = 'Microsoft'
The above query could be rewritten to make use of a view in which

to encapsulate the additional joins, For example:
CREATE VIEW dbo.order_product

AS
SELECT
*
FROM dbo.order AS o
INNER JOIN dbo.product AS p
ON o.product_key = p.product_key
;
UPDATE order_line
SET dispatch_ts = CURRENT_TIMESTAMP
FROM order_product AS op
WHERE order_line.order_key = op.order_key
AND order_line.product_key = p.product_key
AND op.order_no = 'MS12345678-90'
AND op.supplier_nm = 'Microsoft'

Migrating DELETE Statements
Although delete statement syntax can be migrated from SQL Server to
PDW with few if any changes, the performance can be much better
when deleting large percentages of a given large table by converting the
delete statement to a CTAS statement.
Delete statements are executed on a row by row basis and are logged
operations and depending on indexing complexity a lot of database
pages need to be updated and freed. This can cause fragmentation as
well as slow performance to delete large amounts of data due to the
large amount of updates needed.
CTAS statements can provide much better performance on large deletes

since the data to be preserved can be put into a new table while the old
table is dropped.
In the example below the pre 1998 date would be only 20% of say a 10
billion row table or approximately 2 billion rows. In order to delete this
amount of data there would be four possible options.
Option 1 Run a normal delete command which would be a fairly

lengthy operation.
DELETE FROM BigFactTable

WHERE ProductReleaseDate < 1/1/1998
Option 2 If the table is partitioned by year or month then the

appropriate partitions can be switched out to individual tables and
deleted via the ALTER TABLE command syntax. This is a very fast
process but depends on the table already being partitioned
appropriately.
ALTER TABLE BigFactTable SWITCH PARTITION 2 TO

DUMMYTABLE;
DROP TABLE DUMMYTABLE;
Option 3 Convert the delete statement to a CTAS statement by

selecting the data to keep into a new table and then dropping the
old table. This option is not as fast as switching out the partition
but it will be many times faster than the normal delete operation
and the table does not have to be pre-partitioned.
CREATE TABLE BigFactTable_NEW

WITH (DISTRIBUTION = HASH (ID))
AS
SELECT
*
FROM BigFactTable
WHERE ProductReleaseDate >= 1/1/1998
RENAME OBJECT BigFactTable TO BigFactTable_OLD

RENAME OBJECT BigFactTable_NEW TO BigFactTable

DROP TABLE BigFactTable
-- The original table is distributed on a

-- column named ID.
-- The same distribution column is maintained
-- in the new table.
Option 4 Execute a Trickle Delete. Which executes the delete

statement from within a loop, deleting a small number of rows,
approximately 100,000 each iteration. Reducing any potential
logging overhead.
DECLARE @count INT = 1;
WHILE @count > 0

BEGIN
IF OBJECT_ID('tempdb..#DeleteInc') IS NOT NULL
DROP TABLE #DeleteInc
CREATE TABLE #DeleteInc

WITH (LOCATION=USER_DB, DISTRIBUTION=REPLICATE)
AS
SELECT TOP 100000
ID
FROM dbo.BigFactTable
WHERE ProductReleaseDate < 1/1/1998
SET @count = (SELECT COUNT(*) FROM #DeleteInc)
IF @count > 0
BEGIN
DELETE FROM dbo.BigFactTable
WHERE ID IN (SELECT ID FROM #DeleteInc)
END
END
@@ROWCOUNT Workaround
PDW currently does not support @@ROWCOUNT or ROWCOUNT_BIG
functions. If you need to obtain the number of rows affected by the last
INSERT, UPDATE or DELETE statement, you can use the following SQL:
-- @@ROWCOUNT
SELECT CASE WHEN MAX(distribution_id) = -1
THEN SUM(DISTINCT row_count)
ELSE SUM(row_count)
END AS row_count
FROM sys.dm_pdw_sql_requests
WHERE row_count <> -1
AND request_id IN (SELECT TOP 1
request_id
FROM sys.dm_pdw_exec_requests
WHERE session_id = SESSION_ID()
ORDER BY end_time DESC)

Stored Procedures
Stored Procedure creation is supported in PDW and supports most
features available in SQL Server. As opposed to SQL Server however the
individual commands in PDW stored procedures execute in parallel
across all the nodes rather than serially like they do in SQL Server.
SQL Server PDW supports nesting of Stored Procedures up to 8 levels

deep compared to the 32 levels deep supported by SQL Server.
The following stored procedure features are not supported in SQL Server
PDW:
Temporary stored procedures
Numbered stored procedures
Table valued parameters for stored procedures
Extended stored procedures
CLR stored procedures
Read-only parameters for stored procedures
Encrypted stored procedures
Execution of stored procedures under a different user context (the

EXECUTE AS clause)
Default parameters
The RETURN statement
Other Functionality Limitations and Differences

The following functionality is not supported in PDW at this time.
Functions UDF and CLR Currently Not Supported
Rules Rule Syntax is not supported. However defaults and

constraints can be defined on tables. See the CREATE TABLE
statement.
Cursors Not supported at this time. Cursor operations are by

definition designed to process data sets in a row by row
fashion rather than in a set or batch which is the optimal
way to process large amounts of data. Cursors operations
are not recommended in enterprise data warehouse
solutions.

Functions UDF and CLR Currently Not Supported
Optimizer Hints Optimizer hints should not be migrated from SQL Server to
PDW.
If code is to be migrated from SQL Server query hints can be

removed manually via a code review process.
Optimizer Hints vary significantly between PDW and SQL

Server Many of the SQL Server hints do not apply and
indeed if migrated could force poor performance. The
distributed execution plan that is created and executed on
PDW can make hints incompatible.
In PDW hints can be used to specify the join type as well as

to specify if a table is to be replicated or distributed in order
to resolve an incompatible query join situation. It is
recommended that the use of hints is avoided unless
adequate and ongoing testing verifies that hints should be
used.
Statistics vs. Optimizer Hints
Up to date table statistics enable the optimizer to made the

best evaluation and pick the most appropriate execution
plan, Join order and join types. Conversely Invalid or old
statistics can guide the optimizer to pick a sub optimal
execution plan. Query hints should only be considered as a
last option and only once statistics are created and up to
date. Table Statistics creation is a performance
optimization topic and covered in the PDW product
documentation.
Security Considerations - Migration

When moving a database from SQL Server to PDW all the metadata of
the dependent entities and objects in the database must be recreated
on the destination platform. This includes the user security access as
well as logins and passwords.
PDW supports Transparent Data Encryption (TDE), allowing for real-time

I/O encryption and decryption of the data and log files. Encrypted data
on SQL Server will first have to be decrypted before being able to be
loaded into PDW, where if TDE has been enabled it will be encrypted
and therefore protected at rest. Certificates and Keys will need to be
recreated.
Syntax Comparisons
CREATE LOGIN
The CREATE LOGIN command is simplified substantially in PDW. Login
creation with a certificate or an asymmetric key is not supported in the
current release.

CREATE LOGIN loginName { WITH <option_list1> | FROM
WINDOWS }
<option_list1> ::=
PASSWORD = { 'password' } [ MUST_CHANGE ]
[ , <option_list2> [ ,... ] ]
<option_list2> ::=
CHECK_EXPIRATION = { ON | OFF}
| CHECK_POLICY = { ON | OFF}
When migrating any database it is important to note that while login

passwords can be easily reset, user logins, user names as well as
database access and permissions on objects must be maintained on the
new database platform. An approach to accurately recreating the logins,
users and roles on PDW is to first retrieve the metadata for these logins,
users and roles on SQL Server together with any associated permissions.
This can be achieved by using the database scripting wizard and
specifying that logins and object level permissions should also be
scripted.
The SQL Server database scripting wizard can be initiated by right

clicking on a user database and selecting Tasks and then Generate
Scripts The scripting wizard will generate the SQL Statements to
recreate the appropriate permissions.
The following T-SQL functions can be used to query the SQL Server
security metadata, allowing you to perform command level validation on
individual users and related object permissions.
NOTE: These functions are not supported by SQL Server PDW.

fn_builtin_permissions()
This function lists all of the securable classes available in SQL Server. A
securable class is any object that can be assigned permissions. For
example tables, views, procedures etc. The values returned will be used
in the execution of the fn_my_permissions() function to list out the
permissions for nominated users on nominated securable objects.
fn_my_permissions()
This lists out the permissions that a given user context has been granted
on a securable object.
For example:
SELECT DISTINCT
class_desc
FROM fn_builtin_permissions(default)
;
Lists out 25 securable objects in SQL Server 2012.
The example below lists the permissions that user "danny" has on the
object "table_1". The object could be a table, view, procedure etc. Since
the "fn_my_permissions()" function is executed in the current user
context the user context must be set and reset as needed. The user must
exist in the database.
EXECUTE AS USER = 'danny';

SELECT
entity_name
, permission_name
FROM fn_my_permissions('testdb.dbo.table_1',
'OBJECT')
ORDER BY subentity_name
, permission_name
;
REVERT;
GO
This example lists the permissions that user "danny" has on the database
called "testdb".
EXECUTE AS USER = 'danny';

SELECT
entity_name
, permission_name
FROM fn_my_permissions('testdb', 'DATABASE')
ORDER BY subentity_name
, permission_name
;
REVERT;
GO

This migration guide covers the differences between SQL Server and SQL
Conclusion Server Parallel Data Warehouse (PDW), discussing the steps necessary to
migrate a SQL Server database to SQL Server PDW.
Analytics Platform System Website http://www.microsoft.com/aps/

For More SQL Server Website - http://www.microsoft.com/sqlserver/
Information SQL Server TechCenter - http://technet.microsoft.com/en-us/sqlserver/
SQL Server DevCenter - http://msdn.microsoft.com/en-us/sqlserver/
Did this paper help you? Please give us your feedback. Tell us on a scale
Feedback of 1 (poor) to 5 (excellent), how would you rate this paper and why have
you given it this rating? For example:
Are you rating it high due to having good examples, excellent

screenshots, clear writing, or another reason?
Are you rating it low due to poor examples, fuzzy screenshots,

unclear writing?
This feedback will help us improve the quality of the white papers we
release.
Send feedback.

Appendix - Unsupported SQL Server Syntax
Appendix The following Syntax is not directly supported.
Category Command Notes
Aggregate functions CHECKSUM_AGG Not Applicable in PDW.
Aggregate functions GROUPING Not Applicable in PDW.
Aggregate functions GROUPING_ID Not Applicable in PDW.
ALTER ALTER APPLICATION ROLE Not Applicable in PDW.
ALTER ALTER ASSEMBLY Not Applicable in PDW.
ALTER ALTER ASYMMETRIC KEY Not Applicable in PDW.
ALTER ALTER BROKER PRIORITY Not Applicable in PDW.
ALTER ALTER CREDENTIAL Not Applicable in PDW.
ALTER ALTER CRYPTOGRAPHIC PROVIDER Not Applicable in PDW.
ALTER ALTER DATABASE AUDIT SPECIFICATION Not Applicable in PDW.
ALTER ALTER DATABASE Compatibility Level Not Applicable in PDW. Compatibility level in PDW has been set to
emulate SQL Server so that third party applications continue to work.
ALTER ALTER DATABASE Database Mirroring Not Applicable in PDW.

High availability is provided by appliance infrastructure and database
mirroring is not utilized.
ALTER ALTER DATABASE File and Filegroup Options Not Applicable in PDW.
Files and Filegroups are automatically defined by PDW and no user
intervention is required.
ALTER ALTER ENDPOINT Not Applicable in PDW.
ALTER ALTER EVENT SESSION Not Applicable in PDW.
ALTER ALTER FULLTEXT CATALOG Not Applicable in PDW.
ALTER ALTER FULLTEXT INDEX Not Applicable in PDW.
ALTER ALTER FULLTEXT STOPLIST Not Applicable in PDW.
ALTER ALTER FUNCTION Not Applicable in PDW.
ALTER ALTER MESSAGE TYPE Not Applicable in PDW.
ALTER ALTER PARTITION FUNCTION Not Applicable in PDW.

ALTER ALTER PARTITION SCHEME Not Applicable in PDW.
ALTER ALTER QUEUE Not Applicable in PDW.
ALTER ALTER REMOTE SERVICE BINDING Not Applicable in PDW.
ALTER ALTER RESOURCE GOVERNOR Not Applicable in PDW.

Resource governance is already implemented via workload
management resource classes in PDW. PDW simplifies the syntax
and complexity to allow resource governance during large batch
execution. See books online for more information under the
heading Workload Management.
ALTER ALTER RESOURCE POOL Not Applicable in PDW.

See books online for more information under the heading Workload
Management.
ALTER ALTER ROUTE Not Applicable in PDW.
ALTER ALTER SERVER AUDIT Not Applicable in PDW.
ALTER ALTER SERVER AUDIT SPECIFICATION Not Applicable in PDW.
ALTER ALTER SERVICE Not Applicable in PDW.
ALTER ALTER SERVICE MASTER KEY Not Applicable in PDW.
ALTER ALTER SYMMETRIC KEY Not Applicable in PDW.
ALTER ALTER TRIGGER Not Applicable in PDW.

Triggers are not supported in PDW.
ALTER ALTER VIEW Not Applicable in PDW.

DROP the view and re-CREATE the view instead.
ALTER ALTER WORKLOAD GROUP Not Applicable in PDW.

Management.
ALTER ALTER XML SCHEMA COLLECTION Not Applicable in PDW.

XML Data types are not supported in PDW.
BACKUP / RESTORE BACKUP MASTER KEY Not Applicable in PDW.
BACKUP / RESTORE BACKUP SERVICE MASTER KEY Use BACKUP CERTIFICATE instead.
BACKUP / RESTORE RESTORE FILELISTONLY Not Applicable in PDW.
BACKUP / RESTORE RESTORE LABELONLY Not Applicable in PDW.
BACKUP / RESTORE RESTORE MASTER KEY Not Applicable in PDW.
BACKUP / RESTORE RESTORE REWINDONLY Not Applicable in PDW.
BACKUP / RESTORE RESTORE SERVICE MASTER KEY Not Applicable in PDW.
BACKUP / RESTORE RESTORE VERIFYONLY Not Applicable in PDW.
Clause OUTPUT Clause Not Applicable in PDW.
Clause WITH common_table_expression Non-Recursive form is supported.

Collation COLLATIONPROPERTY Not Applicable in PDW.
Collation Functions / Operators COLLATE Not Applicable in PDW.
Collation Functions / Operators SQL Server Collation Name Not Applicable in PDW.
Collation Functions / Operators Windows Collation Name Not Applicable in PDW.
Configuration Functions @@DBTS Not Applicable in PDW.
Configuration Functions @@LANGID Not Applicable in PDW.
Configuration Functions @@LOCK_TIMEOUT Not Applicable in PDW.
Configuration Functions @@MAX_CONNECTIONS Not Applicable in PDW.
Configuration Functions @@MAX_PRECISION Not Applicable in PDW.
Configuration Functions @@NESTLEVEL Not Applicable in PDW.
Configuration Functions @@OPTIONS Not Applicable in PDW.
Configuration Functions @@REMSERVER Not Applicable in PDW.
Configuration Functions @@SERVERNAME Not Applicable in PDW.
Configuration Functions @@SERVICENAME Not Applicable in PDW.
Configuration Functions @@TEXTSIZE Not Applicable in PDW.
Control of Flow CONTINUE Not Applicable in PDW.
Control of Flow GOTO label Not Applicable in PDW.
Control of Flow RETURN Not Applicable in PDW.
Control of Flow WAITFOR Not Applicable in PDW.
CREATE CREATE AGGREGATE Not Applicable in PDW.
CREATE CREATE APPLICATION ROLE Not Applicable in PDW.
CREATE CREATE ASSEMBLY Not Applicable in PDW.
CREATE CREATE ASYMMETRIC KEY Not Applicable in PDW.
CREATE CREATE BROKER PRIORITY Not Applicable in PDW.

Service Broker functionality is not supported in PDW.
CREATE CREATE CONTRACT Not Applicable in PDW.
CREATE CREATE CREDENTIAL Not Applicable in PDW.
CREATE CREATE CRYPTOGRAPHIC PROVIDER Not Applicable in PDW.
CREATE CREATE DATABASE AUDIT SPECIFICATION Not Applicable in PDW.
CREATE CREATE DEFAULT Not Applicable in PDW.
CREATE CREATE ENDPOINT Not Applicable in PDW.
CREATE CREATE EVENT NOTIFICATION Not Applicable in PDW.
CREATE CREATE EVENT SESSION Not Applicable in PDW.
CREATE CREATE FULLTEXT CATALOG Not Applicable in PDW.
CREATE CREATE FULLTEXT INDEX Not Applicable in PDW.

CREATE CREATE FULLTEXT STOPLIST Not Applicable in PDW.
CREATE CREATE FUNCTION Not Applicable in PDW.
CREATE CREATE MESSAGE TYPE Not Applicable in PDW.
CREATE CREATE PARTITION FUNCTION Not Applicable in PDW.
CREATE CREATE PARTITION SCHEME Not Applicable in PDW.
CREATE CREATE QUEUE Not Applicable in PDW.

CREATE CREATE REMOTE SERVICE BINDING Not Applicable in PDW.

CREATE CREATE RESOURCE POOL Not Applicable in PDW.

See Workload management is PDW Books Online.
CREATE CREATE ROUTE Not Applicable in PDW.
CREATE CREATE RULE Not Applicable in PDW.
CREATE CREATE SERVER AUDIT Not Applicable in PDW.
CREATE CREATE SERVER AUDIT SPECIFICATION Not Applicable in PDW.
CREATE CREATE SERVICE Not Applicable in PDW.
CREATE CREATE SPATIAL INDEX Not Applicable in PDW.
CREATE CREATE SYMMETRIC KEY Not Applicable in PDW.
CREATE CREATE SYNONYM Not Applicable in PDW.
CREATE CREATE TRIGGER Not Applicable in PDW.
CREATE CREATE TYPE Not Applicable in PDW.
CREATE CREATE WORKLOAD GROUP Not Applicable in PDW.

Management.
CREATE CREATE XML INDEX Not Applicable in PDW.

CREATE CREATE XML SCHEMA COLLECTION Not Applicable in PDW.

Cryptographic Functions ASYMKEY_ID Not Applicable in PDW.
Cryptographic Functions ASYMKEYPROPERTY Not Applicable in PDW.
Cryptographic Functions CERT_ID Not Applicable in PDW.
Cryptographic Functions CERTPROPERTY Not Applicable in PDW.
Cryptographic Functions CRYPT_GEN_RANDOM Not Applicable in PDW.
Cryptographic Functions DECRYPTBYASMKEY Not Applicable in PDW.
Cryptographic Functions DECRYPTBYCERT Not Applicable in PDW.
Cryptographic Functions DECRYPTBYKEY Not Applicable in PDW.
Cryptographic Functions DECRYPTBYKEYAUTOCERT Not Applicable in PDW.

Cryptographic Functions DECRYPTBYPASSPHRASE Not Applicable in PDW.
Cryptographic Functions ENCRYPTBYASMKEY Not Applicable in PDW.
Cryptographic Functions ENCRYPTBYCERT Not Applicable in PDW.
Cryptographic Functions ENCRYPTBYKEY Not Applicable in PDW.
Cryptographic Functions ENCRYPTBYPASSPHRASE Not Applicable in PDW.
Cryptographic Functions IS_OBJECTSIGNED Not Applicable in PDW.
Cryptographic Functions KEY_GUID Not Applicable in PDW.
Cryptographic Functions KEY_ID Not Applicable in PDW.
Cryptographic Functions SIGNBYASMKEY Not Applicable in PDW.
Cryptographic Functions SIGNBYCERT Not Applicable in PDW.
Cryptographic Functions SYMKEYPROPERTY Not Applicable in PDW.
Cryptographic Functions VERIFYSIGNEDBYASMKEY Not Applicable in PDW.
Cryptographic Functions VERIFYSIGNEDBYCERT Not Applicable in PDW.
Cursor @@CURSOR_ROWS Not Applicable in PDW.
Cursor @@FETCH_STATUS Not Applicable in PDW.
Cursor CURSOR_STATUS Not Applicable in PDW.

Cursors are not supported in PDW.
Cursor Operations @@CURSOR_ROWS Not Applicable in PDW.

Cursor Operations @@FETCH_STATUS Not Applicable in PDW.

Cursor Operations CLOSE Not Applicable in PDW.

Cursor Operations CURSOR_STATUS Not Applicable in PDW.

Cursor Operations DEALLOCATE Not Applicable in PDW.

Cursor Operations DECLARE @local_variable Not Applicable in PDW.

Cursor Operations DECLARE CURSOR Not Applicable in PDW.

Cursor Operations FETCH Not Applicable in PDW.

Cursor Operations OPEN Not Applicable in PDW.

Cursor Operations SET Not Applicable in PDW.


Cursor Operations sp_cursor_list Not Applicable in PDW.

Cursor Operations sp_describe_cursor Not Applicable in PDW.

Cursor Operations sp_describe_cursor_columns Not Applicable in PDW.

Cursor Operations sp_describe_cursor_tables Not Applicable in PDW.

Data Type Functions IDENT_CURRENT Not Applicable in PDW.
Data Type Functions IDENT_INCR Not Applicable in PDW.
Data Type Functions IDENT_SEED Not Applicable in PDW.
Data Type Functions IDENTITY Not Applicable in PDW.

Identity Fields not supported in PDW.
Data Types cursor Not Applicable in PDW.

Data Types hierarchyid Not Supported in PDW.
Data Types image Not Supported in PDW.
Data Types ntext Not Supported in PDW.
Data Types numeric Not Applicable in PDW.

Use DECIMAL instead
Data Types sql_variant Not Supported in PDW.
Data Types table Not Supported in PDW.

Use temp tables instead.
Data Types text Not Supported in PDW.
Data Types uniqueidentifier Not Supported in PDW.
Data Types xml Not Supported in PDW.
DBCC DBCC CHECKALLOC Not Applicable in PDW.
DBCC DBCC CHECKCATALOG Not Applicable in PDW.
DBCC DBCC CHECKCONSTRAINTS Not Applicable in PDW.
DBCC DBCC CHECKDB Not Applicable in PDW.
DBCC DBCC CHECKFILEGROUP Not Applicable in PDW.
DBCC DBCC CHECKIDENT Not Applicable in PDW.
DBCC DBCC CHECKTABLE Not Applicable in PDW.
DBCC DBCC CLEANTABLE Not Applicable in PDW.
DBCC DBCC DBREINDEX Not Applicable in PDW.
DBCC DBCC dllname (FREE) Not Applicable in PDW.
DBCC DBCC DROPCLEANBUFFERS Not Applicable in PDW.

DBCC DBCC FREESESSIONCACHE Not Applicable in PDW.
DBCC DBCC FREESYSTEMCACHE Not Applicable in PDW.
DBCC DBCC HELP Not Applicable in PDW.
DBCC DBCC INDEXDEFRAG Not Applicable in PDW.
DBCC DBCC INPUTBUFFER Not Applicable in PDW.
DBCC DBCC OPENTRAN Not Applicable in PDW.
DBCC DBCC OUTPUTBUFFER Not Applicable in PDW.
DBCC DBCC PROCCACHE Not Applicable in PDW.
DBCC DBCC SHOWCONTIG Not Applicable in PDW.
DBCC DBCC SHRINKDATABASE Not Applicable in PDW.
DBCC DBCC SHRINKFILE Not Applicable in PDW.
DBCC DBCC SQLPERF Not Applicable in PDW.
DBCC DBCC TRACEOFF Not Applicable in PDW.
DBCC DBCC TRACEON Not Applicable in PDW.
DBCC DBCC TRACESTATUS Not Applicable in PDW.
DBCC DBCC UPDATEUSAGE Not Applicable in PDW.
DBCC DBCC USEROPTIONS Not Applicable in PDW.
DROP DROP AGGREGATE Not Applicable in PDW.
DROP DROP APPLICATION ROLE Not Applicable in PDW.
DROP DROP ASSEMBLY Not Applicable in PDW.
DROP DROP ASYMMETRIC KEY Not Applicable in PDW.
DROP DROP BROKER PRIORITY Not Applicable in PDW.
DROP DROP CONTRACT Not Applicable in PDW.
DROP DROP CREDENTIAL Not Applicable in PDW.
DROP DROP CRYPTOGRAPHIC PROVIDER Not Applicable in PDW.
DROP DROP DATABASE AUDIT SPECIFICATION Not Applicable in PDW.
DROP DROP DEFAULT Not Applicable in PDW.
DROP DROP ENDPOINT Not Applicable in PDW.
DROP DROP EVENT NOTIFICATION Not Applicable in PDW.
DROP DROP EVENT SESSION Not Applicable in PDW.
DROP DROP FULLTEXT CATALOG Not Applicable in PDW.

Full text catalogs are not supported in PDW.
DROP DROP FULLTEXT INDEX Not Applicable in PDW.


DROP DROP FULLTEXT STOPLIST Not Applicable in PDW.

DROP DROP FUNCTION Not Applicable in PDW.

Custom declared functions are not supported in PDW
DROP DROP MESSAGE TYPE Not Applicable in PDW.
DROP DROP PARTITION FUNCTION Not Applicable in PDW.

Specify partition using WHERE clause.
DROP DROP PARTITION SCHEME Not Applicable in PDW.
DROP DROP QUEUE Not Applicable in PDW.

Service Broker is not supported.
DROP DROP REMOTE SERVICE BINDING Not Applicable in PDW.

DROP DROP RESOURCE POOL Not Applicable in PDW.

See Workload management is PDW Books Online.
DROP DROP ROUTE Not Applicable in PDW.

DROP DROP RULE Not Applicable in PDW.
DROP DROP SERVER AUDIT Not Applicable in PDW.
DROP DROP SERVER AUDIT SPECIFICATION Not Applicable in PDW.
DROP DROP SERVICE Not Applicable in PDW.
DROP DROP SIGNATURE Not Applicable in PDW.
DROP DROP SYMMETRIC KEY Not Applicable in PDW.
DROP DROP SYNONYM Not Applicable in PDW.
DROP DROP TRIGGER Not Applicable in PDW.

Trigger functionality is not supported in PDW.
DROP DROP TYPE Not Applicable in PDW.
DROP DROP WORKLOAD GROUP Not Applicable in PDW.
DROP DROP XML SCHEMA COLLECTION Not Applicable in PDW.
MANAGEMENT CHECKPOINT Not Applicable in PDW.
MANAGEMENT KILL QUERY NOTIFICATION SUBSCRIPTION Not Applicable in PDW.
MANAGEMENT KILL STATS JOB Not Applicable in PDW.
MANAGEMENT RECONFIGURE Not Applicable in PDW.
MANAGEMENT SHUTDOWN Not Applicable in PDW.
Mathematical Functions RAND Not Applicable in PDW.
Metadata Functions @@PROCID Not Applicable in PDW.
Metadata Functions APPLOCK_MODE Not Applicable in PDW.
Metadata Functions APPLOCK_TEST Not Applicable in PDW.

Metadata Functions ASSEMBLYPROPERTY Not Applicable in PDW.
Metadata Functions COL_LENGTH Not Applicable in PDW.
Metadata Functions COLUMNPROPERTY Not Applicable in PDW.
Metadata Functions DATABASE_PRINCIPAL_ID Not Applicable in PDW.
Metadata Functions DATABASEPROPERTY Not Applicable in PDW.

Use DATABASEPROPERTYEX function instead.
Metadata Functions FILE_ID Not Applicable in PDW.
Metadata Functions FILE_IDEX Not Applicable in PDW.
Metadata Functions FILE_NAME Not Applicable in PDW.
Metadata Functions FILEGROUP_ID Not Applicable in PDW.
Metadata Functions FILEGROUP_NAME Not Applicable in PDW.
Metadata Functions FILEGROUPPROPERTY Not Applicable in PDW.
Metadata Functions FILEPROPERTY Not Applicable in PDW.
Metadata Functions fn_listextendedproperty Not Applicable in PDW.
Metadata Functions FULLTEXTCATALOGPROPERTY Not Applicable in PDW.
Metadata Functions FULLTEXTSERVICEPROPERTY Not Applicable in PDW.
Metadata Functions INDEX_COL Not Applicable in PDW.
Metadata Functions INDEXPROPERTY Not Applicable in PDW.
Metadata Functions KEY_NAME Not Applicable in PDW.
Metadata Functions OBJECT_DEFINITION Not Applicable in PDW.
Metadata Functions OBJECT_SCHEMA_NAME Not Applicable in PDW.
Metadata Functions ORIGINAL_DB_NAME Not Applicable in PDW.
Metadata Functions SCOPE_IDENTITY Not Applicable in PDW.
Metadata Functions STATS_DATE Not Applicable in PDW.
ODBC Scalar Functions OCTET_LENGTH Partial Implementation in PDW.
ODBC Scalar Functions TRUNCATE Partial Implementation in PDW.
Operators (Compound) %= (Modulo EQUALS) (T-SQL) Not Applicable in PDW.

Use expanded syntax instead
Operators (Compound) &= (Bitwise AND EQUALS) (T-SQL) Not Applicable in PDW.
Operators (Compound) *= (Multiply EQUALS) (T-SQL) Not Applicable in PDW.

Operators (Compound) /= (Divide EQUALS) (T-SQL) Not Applicable in PDW.

Operators (Compound) ^= (Bitwise Exclusive OR EQUALS) (T-SQL) Not Applicable in PDW.


Operators (Compound) |= (Bitwise OR EQUALS) (T-SQL) Not Applicable in PDW.

Operators (Compound) += (Add EQUALS) (T-SQL) Not Applicable in PDW.

Operators (Compound) -= (Subtract EQUALS) (T-SQL) Not Applicable in PDW.

Operators (Logical) ALL Not Applicable in PDW.
Operators (Logical) ANY Not Applicable in PDW.
Operators (Logical) SOME Not Applicable in PDW.
Operators (Set) EXCEPT and INTERSECT Not Applicable in PDW.
Operators (String Concatenation) += (String Concatenation) Not Applicable in PDW.
OTHERS BULK INSERT Not Applicable in PDW.
OTHERS DISABLE TRIGGER Not Applicable in PDW.

OTHERS ENABLE TRIGGER Not Applicable in PDW.

Partition $PARTITION Not Applicable in PDW.
Predicates CONTAINS Not Applicable in PDW.
Predicates FREETEXT Not Applicable in PDW.

Full text search is not supported in PDW
Replication PUBLISHINGSERVERNAME Not Applicable in PDW.
Rowset Functions CONTAINSTABLE Not Applicable in PDW.
Rowset Functions FREETEXTTABLE Not Applicable in PDW.
Rowset Functions OPENDATASOURCE Not Applicable in PDW.
Rowset Functions OPENQUERY Not Applicable in PDW.
Rowset Functions OPENROWSET Not Applicable in PDW.
Rowset Functions OPENXML Not Applicable in PDW.
SECURITY ADD SIGNATGURE Not Applicable in PDW.
Security ADD SIGNATURE Not Applicable in PDW.
SECURITY CLOSE SYMMETRIC KEY Not Applicable in PDW.
SECURITY OPEN SYMMETRIC KEY Not Applicable in PDW.
Security Functions HAS_PERMS_BY_NAME Not Applicable in PDW.
Security Functions IS_MEMBER Not Applicable in PDW.
Security Functions IS_SRVROLEMEMBER Not Applicable in PDW.
Security Functions LOGINPROPERTY Not Applicable in PDW.
Security Functions ORIGINAL_LOGIN Not Applicable in PDW.

Security Functions PERMISSIONS Not Applicable in PDW.
Security Functions PWDCOMPARE Not Applicable in PDW.
Security Functions PWDENCRYPT Not Applicable in PDW.
Security Functions REVERT Not Applicable in PDW.
Security Functions SETUSER Not Applicable in PDW.
Security Functions SUSER_ID Not Applicable in PDW.
Security Functions SUSER_SID Not Applicable in PDW.
Security Functions SUSER_SNAME Not Applicable in PDW.
Security Functions sys.fn_builtin_permissions Not Applicable in PDW.
Security Functions sys.fn_my_permissions Not Applicable in PDW.
Security Functions USER_ID Not Applicable in PDW.
Service Broker Statements BEGIN CONVERSATION TIMER Not Applicable in PDW.

Service Broker is not supported
Service Broker Statements BEGIN DIALOG CONVERSATION Not Applicable in PDW.

Service Broker Statements END CONVERSATION Not Applicable in PDW.

Service Broker Statements GET CONVERSATION GROUP Not Applicable in PDW.

Service Broker Statements GET TRANSMISSION_STATUS Not Applicable in PDW.

Service Broker Statements MOVE CONVERSATION Not Applicable in PDW.

Service Broker Statements RECEIVE Not Applicable in PDW.

Service Broker Statements SEND Not Applicable in PDW.

SET Statements SET CURSOR_CLOSE_ON_COMMIT Not Applicable in PDW.
SET Statements SET DEADLOCK_PRIORITY Not Applicable in PDW.
SET Statements SET FIPS_FLAGGER Not Applicable in PDW.
SET Statements SET FORCEPLAN Not Applicable in PDW.
SET Statements SET IDENTITY_INSERT Not Applicable in PDW.
SET Statements SET NOCOUNT Not Applicable in PDW.
SET Statements SET NOEXEC Not Applicable in PDW.
SET Statements SET OFFSETS Not Applicable in PDW.
SET Statements SET PARSEONLY Not Applicable in PDW.
SET Statements SET QUERY_GOVERNOR_COST_LIMIT Not Applicable in PDW.

SET Statements SET REMOTE_PROC_TRANSACTIONS Not Applicable in PDW.
SET Statements SET SHOWPLAN_ALL Not Applicable in PDW.
SET Statements SET SHOWPLAN_TEXT Not Applicable in PDW.
SET Statements SET SHOWPLAN_XML Not Applicable in PDW.
SET Statements SET STATISTICS IO Not Applicable in PDW.
SET Statements SET STATISTICS PROFILE Not Applicable in PDW.
SET Statements SET STATISTICS TIME Not Applicable in PDW.
SET Statements SET STATISTICS XML Not Applicable in PDW.
SQL Server Utilities Statements \ (Backslash) (T-SQL) Not Applicable in PDW.
Statements MERGE Not Applicable in PDW.

See DWloader.exe for more information about loading data into
PDW.
System Functions @@IDENTITY Not Applicable in PDW.
System Functions @@ROWCOUNT Not Applicable in PDW.

A workaround has been provided in a previous section.
System Functions APP_NAME Not Applicable in PDW.
System Functions BINARY_CHECKSUM Not Applicable in PDW.
System Functions CHECKSUM Not Applicable in PDW.
System Functions COLUMNS_UPDATED Not Applicable in PDW.
System Functions CONNECTIONPROPERTY Not Applicable in PDW.
System Functions CONTEXT_INFO Not Applicable in PDW.
System Functions CURRENT_REQUEST_ID Not Applicable in PDW.
System Functions ERROR_LINE Not Applicable in PDW.
System Functions fn_helpcollations Not Applicable in PDW.
System Functions fn_servershareddrives Not Applicable in PDW.
System Functions fn_virtualfilestats Not Applicable in PDW.
System Functions FORMATMESSAGE Not Applicable in PDW.
System Functions GET_FILESTREAM_TRANSACTION_CONTEXT Not Applicable in PDW.
System Functions GETANSINULL Not Applicable in PDW.
System Functions HOST_ID Not Applicable in PDW.
System Functions HOST_NAME Not Applicable in PDW.
System Functions MIN_ACTIVE_ROWVERSION Not Applicable in PDW.
System Functions NEWID Not Applicable in PDW.
System Functions NEWSEQUENTIALID Not Applicable in PDW.
System Functions PATHNAME Not Applicable in PDW.

System Functions ROWCOUNT_BIG Not Applicable in PDW.

A workaround has been provided in a previous section.
System Functions SESSIONPROPERTY Not Applicable in PDW.
System Functions sys.dm_db_index_physical_stats Not Applicable in PDW.
System Statistical Functions @@CONNECTIONS Not Applicable in PDW.

See section in this paper relating to partitions.
System Statistical Functions @@CPU_BUSY Not Applicable in PDW.
System Statistical Functions @@IDLE Not Applicable in PDW.
System Statistical Functions @@IO_BUSY Not Applicable in PDW.
System Statistical Functions @@PACK_RECEIVED Not Applicable in PDW.
System Statistical Functions @@PACK_SENT Not Applicable in PDW.
System Statistical Functions @@PACKET_ERRORS Not Applicable in PDW.
System Statistical Functions @@TIMETICKS Not Applicable in PDW.
System Statistical Functions @@TOTAL_ERRORS Not Applicable in PDW.
System Statistical Functions @@TOTAL_READ Not Applicable in PDW.
System Statistical Functions @@TOTAL_WRITE Not Applicable in PDW.
System Statistical Functions fn_virtualfilestats Not Applicable in PDW.
Text & Image Functions TEXTPTR Not Applicable in PDW.
Text & Image Functions TEXTVALID Not Applicable in PDW.
Transaction Statements BEGIN DISTRIBUTED TRANSACTION Not Applicable in PDW.
Transaction Statements SAVE TRANSACTION Not Applicable in PDW.
Trigger Functions COLUMNS_UPDATED Not Applicable in PDW.
Trigger Functions EVENTDATA Not Applicable in PDW.
Trigger Functions TRIGGER_NESTLEVEL Not Applicable in PDW.
XML Statements WITH XMLNAMESPACES Not Applicable in PDW.
XML Statements xml_schema_namespace Not Applicable in PDW.

SQL Server To SQL Server PDW: Migration Guide (AU3)

Uploaded by

SQL Server To SQL Server PDW: Migration Guide (AU3)

Uploaded by

SQL Server

6 Differences between SMP and MPP

8 PDW Software Architecture

13 SQL Server to PDW Migration

13 PDW Migration Advisor

14 APS Migration Utility

20 Optimization within PDW

Microsoft SQL Server to SQL Server PDW Migration Guide (AU3)

51 For More Information

Microsoft SQL Server to SQL Server PDW Migration Guide (AU3)

Statement necessary to convert a SQL Server database to Parallel Data

This guide has been updated to reflect the T-SQL improvements

Migrating a data warehouse application from Microsoft SQL Server

This white paper explores the remediation activities and provides

Family of robust and scalable enterprise-wide solution.

Microsoft SQL Server to SQL Server PDW Migration Guide (AU3) 4

Business Intelligence Edition

Parallel Data Warehouse

Microsoft SQL Server to SQL Server PDW Migration Guide (AU3) 5

between server insufficient to meet the business requirements. To overcome

Symmetric Multi-Processing (SMP)

Think of this as a typical single server, multi-core with locally attached

Massively Parallel Processing (MPP)

Shared Nothing Architecture

Microsoft SQL Server to SQL Server PDW Migration Guide (AU3) 6

For a database which follows the shared-nothing architecture, each

Microsoft SQL Server to SQL Server PDW Migration Guide (AU3) 7

Generates the parallel query execution plan and coordinates the

Stores and coordinates metadata and configuration data for all of

Manages SQL Server PDW database authentication and

Tracks hardware and software status.

Microsoft SQL Server to SQL Server PDW Migration Guide (AU3) 8

Processes query operations that require transferring data among

Improves query performance by optimizing data transfer speeds.

Data is loaded in parallel directly from the loading server to the

DMS transfers data from each Compute node directly to the

Using PolyBase, DMS transfers data to and from an external

SQL Server Databases

Windows Domain Name Service (DNS) resolves domain names to

Windows Deployment Service

Microsoft SQL Server to SQL Server PDW Migration Guide (AU3) 9

Windows Server Update Services (WSUS) provides the ability to

PDW Microsoft APS Documentation and Client Tools

Community client tools can be obtained via the following locations:

Microsoft Download Center

Microsoft APS Instructor-Led Training

An online course is available on the Microsoft Virtual Academy at:

Microsoft Premier Support for APS

Before you begin the migration of an existing SQL Server data

Planning make a decision on the migration approach.

Microsoft SQL Server to SQL Server PDW Migration Guide (AU3) 10

for Identify candidate databases for migration

Migration Primary Data Warehouses

Primary Data Marts

Identify candidate objects for migration

Identify candidate data for migration

Archive non-required data

High-Availability & Disaster Recover

Security & Data Encryption

Number of Users & Concurrency

Design Complexities (ELT/ETL, UDTs, UDFs, Physical Storage)

Microsoft SQL Server to SQL Server PDW Migration Guide (AU3) 11

Capacity Planning (current and future storage requirements)

Backup & Restore

Approaches Warehouse, each of which provides different costs and benefits.

The three most widely adopted migration approaches are:

1:1 Migration The existing data model and schema is moved

Redesign The data model is redesigned and the application is re-

Evolution The evolutionary approach is a combination of the

Microsoft SQL Server to SQL Server PDW Migration Guide (AU3) 12

Migration Drivers Migration Approach

Performance 1:1 Migration or Redesign or Evolution

Availability 1:1 Migration or Evolution

Scalability 1:1 Migration or Evolution

Single View of the Business Redesign or Evolution

Business Questions (Complexity) Redesign or Evolution