lun configuration best practice

[sdewar83]

Hi

what is the best practice for lun configuration?

we have a MD32220i.  I was going to create a raid 10 array for our operational vm's and a raid 5 array for our development vm's

how should i create the lun's within those arrays?  just 1 big lun on each? or a lun per vm?  a lun per several vm's?

Advice appreciated

S.

[Josh26 · Accepted solution]

Hi,

Definitely dont' create one LUN per VM - it's just administrative overhead.

There's a big "it depends". If your average VM is 300GB like ours, I recommend placing five on each LUN, so you can create LUNs just a bit below 2TB. With vSphere 5 you can go above the 2TB limit, but just because you can doesn't mean you should.

Remember the more LUNs, the more capacity overhead you have, but you don't necessarily want your entire workload on a single LUN.

[logiboy123]

Like Josh said it depends on

1) Size of disks in SAN

2) Speed of disks in SAN
3) Consolidation ratio required on LUN's

4) Performance requirements on LUN's

My preference is to always use 8MB block size under vSphere 4.x no matter the size of the LUN. Under vSphere 5 it doesn't matter as this is system defined at 1MB.

VMware typically says that a datastore should be fine with roughly 15 VMs and their data disks. So if you find out what your average guest storage requirement is this will help you find the answer, keeping in mind that a 30% margin of free space is recommended for each datastore.

Regards,

Paul

[sdewar83]

Do you know if there is any benefit to having databases on separate LUNs to the vm's accessing them?  

is there any benefit if that lun is on the same raid array anyway?

is there a benefit to having them in a seperate LUN on a separate array?  (i guess if one array blows up you at least either have all your data or all your vm's rather than losing both in one hit...) 

[logiboy123]

That depends on what you are trying to protect against;

1) Datastore failure

2) SAN failure

3) Network failure

4) Application (SQL) failure

Typically the best form of SAN protection is SAN based replication.

Regards,

Paul

[sdewar83]

how about from a pure performance perspective as opposed to -  data loss?

would we get better peformance for our databases by having them on different lun's or disk arrays?   (i guess that might depend upon the raid type that you use?)

we support a system created by someone else and they've created a different lun for each database... i figured there must be some data protection or performance reason they would've done that.

[Techstarts]

sdewar83 wrote:

how about from a pure performance perspective as opposed to -  data loss?

would we get better peformance for our databases by having them on different lun's or disk arrays?   (i guess that might depend upon the raid type that you use?)

we support a system created by someone else and they've created a different lun for each database... i figured there must be some data protection or performance reason they would've done that.

From pure performance perspective do following.

1. Ensure Database files and log files are on different LUN's
2. Ensure Temp DB is also on dedicated LUN
3. Remember log files are sequential and database files are random. Both shouldn't be on same LUN's. Then even if log files or database files of multiple database are on same lun shouldn't be a problem.
4. For better performance consider RAID10

Hope it helps

[sdewar83]

Thanks!

so separate lun's even if they're on the same raid array will increase performance?

or is having the two on different arrays a no-brainer

[Techstarts]

To make it more simpler, do as you would in physical box. In physicalbox we put Logs on different disk and database files on different disk

[sdewar83]

Thanks for your advice, you've been a great help.

If i have to have either the dabatases or the transaction logs on the same disk as VMs.... which should i put there?

I figure the databases as there is more likely to be random writes on the disks with the VMs?  and then I would put the transaction logs which are sequential on their own disk?

[hardcoreit]

You can have C:\ for windows D:\ for Database files and L:\ for logs. Make sure these vmdks are all seperate. you should have 3 vmdks. Then place these vmdks all on seperate luns. That will be best for performance.

[TomHowarth]

So you are saying the each disk should be on its own VMFS LUN why?  the administrative headache that this will cause would be hard to swallow even in a small environment

[vGuy]

For Randomness/Sequentiality of Data access: http://communities.vmware.com/message/1993765#1993765

in essence, the sequential reads/writes may end up enventually as random to the disk..

vSphere assigns specific no. of queues for each device/lun (by default 32)...therefore separate LUNs for database and logs means more queues and enhanced performance..you also have an option to create a high IO virtual disk lun on a separate raid set (raid 10 perhaps).....hth!

[hardcoreit]

Why is it an administrative nightmare? How often do you go in and mess with your luns? Are you constantly changing the size of your vmdks? Do your vms grow that much? Like I said I have had it for years and I never have to change that unless I am increasing the size of a vmdk and that only happens once in a while. And if you did that in a consolidated lun environment you would have to do the same. What other administrative tasks are you referring to?

[hardcoreit]

In a nutshell, I would do this if you are a small environment between 100 and 150 vms and if you use srm and volume based snapshots/replication. That is the only reason I would do this.