vPower NFS Store slooooow

[muhaynes]

Hey guys, I'm hoping you can give me some insight into a way to tweak my setup here to increase speeds in the Virtual Labs.

I'm running a iSCSI SAN running in isolation from my regular network, works great. When I configure the Virtual Lab, I point it at one of my mounted SAN LUNs. Everything works, just not well or fast. We're talking 20MB/s and 30ms of latency on the disks. Obviously my Surebackups take forever, but more importantly it's seriously hurting it's usefulness as a sandbox environment. I mean, these VMs are so slow to start that services timeout on bootup, causing problems one can easily mistake for test failures, etc.

The backup box is Dual Xeon E5640s, 24GBs of RAM, running on a 12x2TB SATA2 RAID 50 array - I don't believe it to the be the problem.

Now I've noticed that the vPower NFS datastore is advertising and being connected to via the LAN Subnet. This is (I'm assuming) why my disk access is so slow. Is there any way to configure it to connect via the SAN subnet?

[Gostev]

With locally stored backups, you should be seeing under 1 minute boot times of VM from backups (those not running heavy-weight applications or domain controllers). Unless your production VMs are heavily fragmented of course. What boot times are you getting?

20MB/s sounds normal for vPower NFS, not sure about 30ms since I never measured, but of course the latency would be much higher than with raw storage because of all on the fly processing (decompression) going in parallel. Still, I've seen customers reporting vPower NFS beat older SANs for boot and file transfer performance.

Could there be random read performance issue with your RAID controller? Of course, you won't mention this during backups, since that is all about writes.

You can manually mount vPower NFS datastore on any subnet (just be sure that the host has VMkernel on that interface), search for "vPower NFS FAQ" topic on this forum - I have listed the step-by-step process there.

Thanks!

[muhaynes]

Thanks Gostev, I will try to move the NFS store to another subnet and see if that helps at all. The random disk access on that array isn't great I'm sure, I do run a secondary software iSCSI target on there for dev/test stuff, last tests I ran on it had it getting about 1500 iops in random workloads.

I'm beginning to lean away the problem being related to the datastore networking, sadly (because that seemed to be the 'easy' fix). Moving a regular VM to that datastore yields low latency and alright throughput, might just be the double-whammy of decompression + slow storage.

[muhaynes]

No love running on the SAN subnet - in fact it got slower somehow. Ah well, back to the drawing board. Thanks again.