Storage
Storage
Storage
Topics
FC SAN (Fiber Channel Storage Area Network) iSCSI (Internet Small Computer System Interface) NAS (Network Attached Storage) VMFS (Virtual Machine File System)
07/04/2013
Components of FC SAN
07/04/2013
Storage System This is the hardware itself, which manages disks into logical units (RAID). LUN Logical Unit Number A part of logical unit. One logical unit can consist of multiple LUNs (and usually does) RAID, JBOD (Just a Bunch of Disks)*, Hot Spare, Parity** SP Storage Processor WWN (World-Wide Name) 64 Bit number, to which we can attach a label Fabric part of SAN storage, components which are part of one access path (serverport, switch, storage port) HBA Host Bus Adapter the card which connects a host to the Fiber Channel network Switch) A non-RAID (SAN disk architecture
**A parity bit is a bit that is added to ensure that the number of bits with the value one in a set of bits is even or odd. Parity bits are used as the simplest form of error detecting code.
07/04/2013
FC / SAN Switch The most important component of the FC network, the HBAs can be connected to the SAN switch
SAN Zoning A method, which makes it possible to create logical groups containing WWN nodes over the physical Fabric Channel network; it is applied on the FC switch level; it is similar to VLANs used in Ethernet networks LUN Masking A method, which makes it possible to hide certain Luns from certain hosts; usually implemented on HBA level
07/04/2013
SAN Zones
07/04/2013
07/04/2013
Zoning
07/04/2013
Storage multipath
Storage multipath is a method which is vital in order to achive and maintain high availablity Active/Active array (rarely used) Active/Passive array (frequently used) There are two types of multipathing policies in an ESX evironment:
Fixed The ESX only uses the preferred path, path failover policy will only be used in case of failure. MRU (Most Recently Used) : The path used most recently will be used in case of a failure, this is set by the ESX automatically. RoundRobin Experimental only
07/04/2013
10
07/04/2013
07/04/2013
12
iSCSI storage
Topics
Components and addressing of iSCSI iSCSI hardware and software initiators Configuration of iSCSI softwareinitiated storage
07/04/2013
14
What is iSCSI
SCSI transport protocol, creating connection to storage devices over stanard TCP/IP protocol It maps SCSI block-based storage over TCP/IP The Initiators, for example an HBA in the ESX, send SCSI commands towards the targets within the iSCSI Storage System
The task itself does not differ from Fiber Storage connection
IP
07/04/2013 15
VMs access to raw iSCSI LUNs VMotion service for VMs which are on iSCSI LUNs
16
07/04/2013
IP Network
* Software implementation
07/04/2013 17
Addressing of IP SAN
iSCSI target name iqn.1992-08.com.netapp:stor1
iSCSI alias stor1 IP address 192.168.36.101
IP Network
07/04/2013
18
Two discovery methods are supported: Static configuration SendTargets iSCSI device sends target info on request
192.168.36.101:3260
SendTargets request
07/04/2013
19
CHAP (Challenge-Handshake Authentication Protocol) Using this authentication method makes it possible to authenticate through the network without actually sending it.
CHAP password: K
CHAP password: K
Log into target (only request) Computes using formula against K using C and sends result R
C R
Create random hash/computation value C send value back to ESX Server; perform local computation against K using C to come up with R. Compares local R against returned R; If match, then ESX Server is authenticated.
07/04/2013
20
07/04/2013
21
Both Service Console and VMKernel has to be able to reach the iSCSI Storage There are two possibilities for this: 1. Service Console and VMKernel port share a virtual switch, in the same subnet.
2. Local routing, so that Service Console and VMKernel both reach the iSCSI storage
07/04/2013
22
07/04/2013
23
07/04/2013
24
07/04/2013
25
The iSCSI name and alias will be filled automatically after activation of the initiator
07/04/2013
26
the IP address of all target can be given on Dynamic Discovery tab All pinging targets will appear on Static Discovery tab
27
07/04/2013
CHAP is turned off by default After enabling it a username and password can be supplied
28
07/04/2013
07/04/2013
29
IP Network
Counts as one network interface NIC teaming and multiple SPs allow for multiple paths
30
07/04/2013
07/04/2013
31
VMFS datastore
Topics
VMFS-3 filesystem Creating VMFS partition VMFS partition align VMFS Metadata VMFS UUID VMFS Extending Distributed Lock Handling Importance of SCSI ID, LUN syncronization, DR site scenario Raw Device Mapping
33
07/04/2013
What is VMFS?
A filesystem optimized for virtualization, capable of clustering In case of shared storage solution, capability of SAN interrupts In service console /vmfs/volumes contains the VMFS volumes
Each virtual machines files are located in its own subdirectory
07/04/2013
34
Creating a VMFS
07/04/2013
35
Creating a VMFS
07/04/2013
36
Misaligned partitons cause performance problems (I/O) Formatting from VI will result in automatic alignment but the recommendations of the manufacturer should be taken into consideration VM Guest OS align (the file system of the guest OS has to be aligned as well)
07/04/2013
37
07/04/2013
38
07/04/2013
39
VMFS properties
07/04/2013
40
VMFS properties
UUID: Universal Unique ID This identifies the VMFS volume Similar to SID (AD), but in this case not the RID master FSMO creates them, but the LVM (Logical Volume Manager) Structure of UUID: 123abc98-12abc456bd12-000569ab56bf
123abc98 Time of VMFS creation or resigniture 12abc456 CPU timestamp counter value Bd12 Random number 000569ab56bf MAC address of service console IF (which created or resignatured the UUID)
41
07/04/2013
ESX1 and ESX2 mount a VMFS volume The host IDs appear in the VMFS heartbeat The host IDs are updated regularly ESX1 locks the file, for identification the host ID is used In case ESX1 stops or loses connection to the volume, the lock will be out of date (the Distributed Lock Manager DLM sees that a dead host owns a file) e.g. HA failover ESX2 tries to lock the file previously locked by ESX1 ESX2 looks up the host ID record of ESX1 in its VMFS heartbeat If the host ID record is not updated by ESX1, ESX2 takes over the lock The other hosts dont try to lock the file agains ESX2 Tries to rollback the VMFS file journal DLM asks the journaling module to recover the filesystem (if necessary) The lock is at the DLM which uses the out-of-date lock The old lock is unregistered, and ESX2 gets to lock the file
42
07/04/2013
VMFS properties
07/04/2013
43
Extend VMFS
The size of a VMFS is dynamically extendable Possible to add more space to a VMFS online Possible to create filesystems bigger than 2 TB
Before spanning After spanning
vmhba0:0:3:1
vmhba0:0:6:1
vmhba0:0:3:1
vmhba0:0:6:1 Extent
07/04/2013
44
Datastore
Datastores consist of VMFS partitions Max. 32 extent (similar to windows) Ideal for creating big datastores Sometimes it might be necessary (the customer is increasing the datastore slowly) If the master is lost, data from all extents are also lost All ESX has to see all extents LUN, otherwise the whole datastore falls apart Because of these the master VMFS can not be on local storage because then not all VM can see the datastore, or if it is so, all VMFS has to be on local storage, and then only one VM can see it
45
07/04/2013
The partiton can not be VMFS to which you want to extend If you choose one which contains data, the program (VI) will notify you that the data will be lost
07/04/2013
46
If one LUN can be maximum 2 TB of size, how is it possible to create a max 64 TB file system? Solution: extending: 32 x 2 TB = 64 TB
07/04/2013
47
07/04/2013
48
SCSI ID
07/04/2013
49
When a VMFS-3 volume is created on a LUN, the SCSI ID comes from the SAN storage (unique) and is stored in the header of the LVM When an ESX scans the LUNs of the storage and finds a VMFS-3 file system, it compares the SCSI ID information of the storage and the one stored in the VMFS LVM metadata This is why it does not mount a replicated LUN, because the SCSI ID is not recognized It can not be mounted because the UUID is the same as the originals and a VM can not mount VMFS volumes which have the same UUID
50
07/04/2013
LVM.EnableResignature LVM.DisallowSnapShotLUN When a LUN is cloned (LUN1 replicated to LUN2) to another storage, the ESX hosts do not find the cloned LUN when rescanning DisallowSnapShotLUN ESX B cannot see LUN1 EnableResignature If ESX B can see LUN1 resignature is necessary, otherwise there will be a conflict because of the UUID This also means that the VMs need to be reregistered under the VI because the UUID is also stored in the vmx files
51
07/04/2013
07/04/2013
52
RDM
RDM Raw Device Mapping Direct link to a raw LUN on a storage Its slightly faster the VMFS (e.g. database) but it depends on the application Physical and virtual compatibility mode Physical: the most direct access for a VM However we lose some options like template creating, snapshoting (VCB backup) migrations which cause moving of disks; these are prerequisites for some services, such as Microsoft Clustering under VMWare Virtual: We have the opportunity to fully use VMWare functions (snapshot, cloning, migrating)
53
07/04/2013
Topics
07/04/2013
55
NFS volumes are the same in FC and iSCSI They are capable of storing the VMs disks
ISO templates VM templates
When the VMs disks are on a NAS, they can be vMotion-ed in case: CPU compatibility All necessary network and storage are reachable on the target as well
56
07/04/2013
NAS
Network Attached Storage Storage shared over the network at a filesystem level
07/04/2013
07/04/2013
58
07/04/2013
59
A seperate IP is necessary for using NAS, different than the Service Console IP
07/04/2013
60
07/04/2013
61
07/04/2013
62
Summary
An NFS datastore can be used to hold templates, ISO images and the files that make up a VM ESX Server supports NFS version 3 over TCP A VMkernel port must be configured in order for the VMkernel to access the IP-based storage
63
07/04/2013