new vmware and equallogic setup

[wb2]

I have two physical vmware hosts attached to an equallogic iscsi san(single array). What would be the best setup on the san side? I'll have 3 to 4 VM's on running from each physical host. Would it be best to chop the storage up in half and have two storage pools? Or just simply have one storage pool and let each host store their VM's on it?

[mike_laspina]

Hi,

More VMFS LUs/stores will be better from a perfomance side. With many VM's on a VMFS store you will have more SCSI reservations on the VMFS metadata files.

10-15 VMs per VMFS store is common.

vExpert 2009

[wb2]

So in my setup, the Equallogic box would contain one single storage pool, then have it split in half into 2 volumes. One volume for esx host1 and the other volume for esx host2. I'll then put my 4 vm's on each host/volume. How does that sound?

I've configured EMCs before with LUNs etc, but these Equallogic boxes don't seem to be configurable in a similar manner.

[malaysiavm]

equallogic is peanut if you had experienced on EMC SAN administration previously. It is design for simplify IT administration :smileysilly:

Craig

vExpert 2009

Malaysia VMware Communities -

[mike_laspina]

Hi,

Yes. And the volumes can serve either hosts VMs, the idea is to prevent to many VM on a single LUN.

vExpert 2009

[wb2]

I appreciate all the responses.

As far as the VMware ESX support. What is the largest volume they would support? Does ESX also support online expansion of the volumes in case we need to increase it's size?

[mike_laspina]

A single VMFS volume segment size maximum is 2TB. The largest VMFS store possible is 64TB using extents (not recommend by most including me)

You cannot perform online LU expansion with a VMFS store. You can only extend it with an additional LU.

vExpert 2009

[christianZ]

When you have only one EQL box then you can't split it in many arrays - you must decide what raid level you choose- R5, R50 or R10) - and then all the disks (- 2 spares) will be configured with it. Now you can configure the "volumes" - these will be seen by your hosts. Your member will be included in "default" storage group (by default). The "storage group" has more importance by more than one member (box).

I'm using EQL here (more than one).

In addition your volumes shouldn't be too large (e.g.300-500 GB) with no more than 5 vms on each.

[aguacero]

We are running EQL here. We've created multiple datastores and on average size is 500 GB (MAX) and approximately around 8-10 VMs which gives you plenty of space for snapshots (vmware snapshots not EQL). In your environment review the current size of the VMs (vmdk files) and then give your self a gap for snapshots per VM. Create 2 datastores (2 volumes on EQL) let's say 400 GB, you can thin provision them if you want. Then assign 4 VM in each datastore keeping in mind the type of application that are IO intensive like Exchange, SQL, etc. that is running on them.

If you found this information useful, please consider awarding points for "Correct" or "Helpful". Thanks!!!

[wb2]

For the smaller servers this answers all my questions, but we have a couple servers that are 2 TB that I want to convert to VMs. With my Equallogic array, what is my best setup given that VMFS maxes out at 2TB volumes? Just create extents? 500GB volume with another 3 or 4 500GB extents?

[mike_laspina]

To go beyond a 2TB fs you will need to stripe the vmdk in the VM's OS.

The largest supported vmdk or RDM size is 2TB with 8k blocks and this is regardless of the extent size.

vExpert 2009

[matuscak]

The other option for large data stores is to run an iSCSI initiator inside the VM that needs the storage and bypass the VMFS level completely. We do this for large file servers and our busy database servers. It also has the advantage that you actually use additional gig interfaces for traffic as opposed to the ESX software initiator.

[eliot]

I tend to stick to 500Gb for VMFS volumes and create as many as required for the Job. Multiple volumes ensure that the IO is spread accross multiple network cards - because for any given volume and host the VMware s/w intiator will only actually use one card.

[RobVM]

EqualLogic has a knowledgebase question,

VMware ESX 3.x one volume or many volumes?

Which I cannot provide a link for as it is in the salesforce.com KB database. The gist of the answer is, more volumes are better. There is a limit with the ESX iSCSI initiator to 32 outstanding I/Os, and extra requests are paused. And starting/stopping VMs or using vMotion require ESX server to have exclusive use of the volume for a short while. But the articel goes on to say that 450-700 Gib per volume for 10-15 Windows 2008 servers is OK, far more than the 4 per volume you suggest.

Like others in this thread it seems that your columes should be around 500 gig, create at least two, maybe three, leave extra space on your array for snaps and data volumes (e.g. a file server).