Link to home
Start Free TrialLog in
Avatar of Justin Perry
Justin PerryFlag for United States of America

asked on

Need advice on current VM disk structure of multiple LUNs spanned within OS

I have a SQL server VM, running in a vSphere 5.1 environment, and the current vdisk structure is a bit complex and I would like to know if it is actually making a positive difference.  The structure was setup by a previous employee and I am just getting my feet wet with the storage array.

The entire VM disk capacity is approximately 1.5TB.  This is a major issue when backing up because of its size. I can rearrange the space to reduce the size but I dont want to disrupt the current structure if it is actually doing some good (performance-wise).

Here is the structure, as best I can convey:

EMC NX4 SAN Structure:
RAID 5 (4+1) disk pool (10k SAS) - LUN A
RAID 10 (6) disk pool (10k SAS) - LUN B

VMDK Structure for SQL server:
1x 300GB vmdk and 1x 240GB vmdk on VMFS datastore 1 [LUN A] (iSCSI to SAN)
1x 300GB vmdk and 1x 240GB vmdk on VMFS datastore 2 [LUN B] (iSCSI to SAN)

OS Disk Structure:
LUN A - Disk 1 - 300GB (250GB partition and 50GB partition) Spanned
LUN A - Disk 2 - 240GB (240GB partition and 10GB partition) Spanned
LUN B - Disk 3 - 300GB (250GB partition and 50GB partition) Spanned
LUN B - Disk 4 - 240GB (225GB partition, 10GB partition, 5GB partition) Spanned

Disk 1 and Disk 3 are Spanned as Drive D:\ 600GB total
Disk 2 and Disk 4 are Spanned as Drive E:\ 480GB total

Its quite a site to look at on the OS side.  Is this really worth it?
Avatar of Andrew Hancock (VMware vExpert PRO / EE Fellow/British Beekeeper)
Andrew Hancock (VMware vExpert PRO / EE Fellow/British Beekeeper)
Flag of United Kingdom of Great Britain and Northern Ireland image

This seems quite an old school setup, confusing a virtual server with physical.

Personally, I would use a single RAID 10 array, with all spindles in a RAID 10 array.

The more disks = more spindles = more performance = more IOPS.

Spanning disks at the OS level, dangerous if a disk gets corrupted or has a snapshot issue, I would abandone that idea, and just create a simple virtual disk, in a new RAID 10 LUN.

I would then create an OS in a single virtual disk, and data virtual disk, on the same array.

BUT, as you do not want to change the structure, what is the major issue with backup?

Also, I would change to Raw Device Mappings on the LUN for SQL database, and leave the OS as virtual disk (vmdk).

1.5TB is large to backup, and I would use SAN based snapshots to backup such a large database, I would not use Agents in the VM, of VMware Snapshot based backup, e.g. Veeam et all.
It entirely depends on your access patterns.
Typically on a SQL server you want the data, logs, and partitions on physically seperate disks since they operate simultaneously.
You have spanned everything together like a mesh, but since it is not a raid stripe at the OS level the impact will not be as bad.
With each disk serving multiple functions your disks may be seeking a lot more and not transferring much as a result.
Avatar of Justin Perry

ASKER

To add to the complexity, the two RAID groups on the storage array are also serving other datastores and I dont have any additional free space on the array to create another storage pool and RAID group.  Also, the space used by this SQL server is more than the capacity of a single VMDK so breaking the spanning isnt feasible at the moment.  

Also, I left out that the OS is on vdisk on a separate NFS datastore.   The databases are on the logical D: drive and the log files are on the E: drive.  Obviously it makes no real difference since both volumes are using the same sets of spindles on the storage array but I guess it makes people feel better if the OS sees it as separate disks or volumes.
so, what's the backup issue? (as you do not want to change!)
Thats quite a conundrum, if you dont have space to make changes then the options are very limited.
As log files can be very small you could start by temporarily flushing, shrinking, and moving them to OS drives just long enough for you to do something with the E: drive.
Hmmm...  Can you setup a temporary iSCSI host and targets on another computer?  Perhaps buy a pair of large harddisks to add to a server or build a server to facilitate operations during your restructuring.  I use StarWind for temporary projects and virtual cluster labs.
The backup issue is timing.  We have a couple different backup solutions; image level and file level.  The image level backup (disk-to-disk) and the subsequent disk-to-tape job run for over 36hrs total.  The others take less (about 15-16).  And Im still only referencing this one SQL server.  I wont even get into the rest of my environment!

I have quotes for a new backup solution but it has not been approved yet.  

I could possibly move the vmdk's to other datastores but there would be a huge performance hit, possibly more than the system can handle.
Hanccocka, forgive my ignorance.  Can you elaborate on the SAN-based snapshot?  

First, let me tell you what products I already have.  Maybe one of them can do this (or already is):

1. EMC Networker 7.6.4
     VCB/VADP for some servers, traditional for others, some use NMM.
     Some stay on disk, some go to disk then to tape, some go directly to tape

2. vmware VDP (new to 5.1)
     Im still toying with this one.  Works great for small servers.
ASKER CERTIFIED SOLUTION
Avatar of Andrew Hancock (VMware vExpert PRO / EE Fellow/British Beekeeper)
Andrew Hancock (VMware vExpert PRO / EE Fellow/British Beekeeper)
Flag of United Kingdom of Great Britain and Northern Ireland image

Link to home
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
Start Free Trial