MetaLUN's, ESX Best Practices and restrictions

[btrotter]

We currently have 4 Fibre Channel LUNS and 2 iSCSI LUN's. We are looking to consolidate the 4 FC LUN's into 1 FC Metalun. We are running 30 VM's, and will probably increase that number a good bit.

I heard that VMWare doesnt recommend more than 15 VM's per LUN. Does this same restriction carry over to Metaluns?

Is there any reason why I shouldnt replace the 4 Fibre Channel LUN's with 1 MetaLUN?

Regards,

Brian

[piacas]

You can use striped MetaLUNs on your EMC backend for ESX storage. Are you consolidating the 4 FC LUNS for managabilty reasons? The problem with Metaluns is the Metahead LUN will become a hot spot on the EMC array as the IO will flow thru it. (thats what I heard) On the plus side, Metaluns will allow you to spread the IOPS over all the disks in the Metalun.

As far as VM's per LUN, Vmware says no more that 32 heavy IO VM's per LUN, or 100 low IO VM's per LUN. That 15 number you mentioned is probably the number of ESX servers attached to the LUN.

[williambishop]

Or you can do up to 300 vm's per lun if you're doing the desktop thing, as long as you have the horsepower on the san side to push it(though it's not really very high in comparison to many things--say a heavy hitting database or the like).

[btrotter]

Great. Thank you for the replies. I feel much better now.

We are looking to consolidate our 4 FC LUN's into 1 FC MetaLUN for manageability, and hopefully performance.

I guess I am thinking that 1 MetaLUN made up of a tray of 300Gb FC drives would offer better performance than 4 seperate LUN's, each with just a few FC drives each.

[jasper9890]

Sorry for reviving this old thread from the dead, but am very curious on the topic and doesn't look like anyone responded.

This statement is curious to me: "I guess I am thinking that 1 MetaLUN made up of a tray of 300Gb FC

drives would offer better performance than 4 seperate LUN's, each with

just a few FC drives each."

You will still hit reservation issues if you are using a single logical datastore won't you, no matter if on the backend it's made up of multiple luns, probably 15 or so disks?