vSphere Datastore Sizing

[ITTech2002]

I have 6 ESX servers that will be hosting 60 Windows Servers 2008 VMs

The average VM or VMDK size is 100GB (1 disk per vm)...  What is the suggested best practice size for my datastores?

I am using iSCSI and Dell Equallogic PS6000 arrays.

[ProPenguin]

I am currently running a 4 ESX server setup with 40 servers.  There harddrives differ in size but it was recommended to me when setting them up to create volumes around 250GBs for best performance.  We are using an Equallogic PS6500 SAN and we have not had any issues running that setup.  Hope this helps.

[ProPenguin]

One last piece of advise, if you decide to create volumes larger than 250GBs, do not create one anywhere close to 2TBs.  If you do it will not show up.  ESX does not support a volume that big.

[Kahonu]

Aloha,

I heard and followed the same advice. No problems.. A while ago I heard that since ESX 4.x iSCSI driver is more efficient, you can feel comfortable using larger LUNS. Is this fact or fiction??

[Josh26]

Hi,

I have always deployed LUNs just under the 2TB limit. I cannot imagine any reason not to.

[idle-jam]

would create a LUN of 600-800GB per LUN. this would ensure that the IOPS are well balanced. DO remember to enable VAAI that woud make the things faster.

http://kb.vmware.com/kb/1021976

[AndreTheGiant]

See also: http://communities.vmware.com/docs/DOC-10990

[Cyberfed27]

It used to be the recommended size for datastores should not exceed 500GB, however with the latest versions of ESX/ESXi that's not really true anymore.

Datastores 1TB or larger are quite common now and perfectly accepted by VMware.

I tend to make mine between 500GB and 1TB just for management purposes.

[ITTech2002]

Last I checked the Max size for a VMFS partision is 2TB - 512Kb

This pertains to ESX 4.1 and this limit will soon change with version 5.0

[ITTech2002]

The volume capacity of a LUN and IOPs available for that LUN/datastore are too separate things, but you bring up a great point. 

Sizing the LUN smaller and allocating less VMs per datastore may reduce the IOPs required per datastore, but it does NOT ensure the datastore has the required the IOPs.  Look at 3PAR, Compellent, and EqualLogic for more details about block level storage.