Getting the most from an Equallogic PS100e

[doggy]

Hi,

We've just got an evaluation Equallogic PS100e. I'm using IOMeter to compare a physical server's hdd response time to that of an Equallogic hosted VM (we P2V'd the same physical host).

The equallogic box has 3 NICs which, apprently, it can use to get 3GB of bandwidth. The ESX server has 6 NICs.

How am I best to confiure the ESX's NICs and the VM's Virtual NICs to get maximum throughput to the Equallogic. Can I team them or anything like that?

Thanks

[dominic7]

You can team them but it won't help your performance. The software iscsi initiator that is built into ESX 3.0.1 won't spread the load across multiple NICs. It will however provide you with failover in the event of a NIC failure.

[woharrow]

To test performance there are a few things you should try to do.

1st use more than one Volume on the array, and attach them all the to server.

if you have 3 NICs on the host in the Vswitch where COS and VMKernel are you should see 3 sessions to the array with each most likely on a different physical NIC as the Vswitch does a round-robin of the connections.

2nd you should make sure that you have at least 3 workers in IOmeter to stress each of the volumes.

The downside is this sort of test really isn't a fair test, DAS is something you can't multi-thread from multiple hosts/guests, while you can do so much more with a SAN like EqualLogic. Sort of like trying to test a chevy with 1 gear against a Porsche with a 6 speed, but don't use anything above 1st gear.....

A more accurate test might be to download Lab Manager and run a ton of VMs against both. (only 1 problem.... you'll need a SAN for the LAB Manager to work which sort of rules out DAS.)

What IOmeter settings are you running?

Thanks

Wade

[doggy]

Thanks Wade.

So, can you confirm my understanding:

We:

1). Install 3 NICs on the VM (whose .vmdk files are hosted on the Equallogic box) and attach it to the same VSwitch as the VMKernel and COS. (do we set any load balancing options in VMWare etc?)

2). We create 3 volumes on the Equallogic box and attach them to the VM

3). We set three IOMeter Workers going, one on each Volume.

Cheers ref the Lab Manager advice, but we do need to compare against DAS. I have been set a mission to show that the Equallogic is capable of hosting our VMs at comparable speed to our current physical setup (mirrored SCSI drives on IBM xSeries).

Am I ever going to be able to get faster IO response times from the Equallogic than from DAS?

IOMeter is setup as:

33% Read 67% write

5MB Transfer request size

We're doing runs on 500MB and 4GB iobw.tst file sizes

Test run for 15 mins.

Thanks

Dom

[woharrow]

you are correct on the setup. with one exception.......

try setting the transfer (block) size to 4KB since this is the default block size in windows. or maybe 8k or 16k. a 5MB block is something I would expect to see with high def video edits, which are large block seq I/O.

also set the queue depth to 16 so it issues some I/O and isn't sitting there sending them one at a time.

if you make these changes and re-run the tests on both your DAS and eql i bet the numbers are very different.

good luck

wade

[christianZ]

Sorry for damaging of your illusions, but as the dominic7 said you won't see any additional throughput when you get more nics - see my answer here:

http://www.vmware.com/community/message.jspa?messageID=672944#672944.

Have you seen this thread here:

http://www.vmware.com/community/thread.jspa?threadID=73745&tstart=0