ESX 3.5 Update 4 terrible read performance

[eschneid]

SAN:
6 x Dell 2950s with 6 15k SAS 450 G drives in RAID 5
4 x Dell 2905s with 6 7.2k SATA 1TB drives in RAID 5
Split between 2 sites.

All volumes are 2-way replicated

All the clients are on Dell m600 blades using Broadcom 5708 NICs at 1 Gig with a dedicated Foundry network. MTU = 1500

Windows 2008 with 2 NICs and MPIO software can write ~90 MB/s and read ~70MB/s constant

CentOS 5.3 with 1 NIC can write ~70 MB/s and read ~70 MB/s constant

**The problem child** I can reproduce this with 4 identical systems in both sites.

VMware 3.5 update 4 with 2 vNICs in a single vSwitch can write ~50 MB/s constant but the reads alternate between long periods (20 sec) of 0 MB/s and short spikes of 50 MB/s.

I have an open ticket with VMware but they haven't provided any real help.

Any ideas?

Thanks,

Eric

[Gauche]

No great ideas without being able to dig deeper. You might want to tail your vmkernel log to see what is going on when there are periods of no IO. Maybe sessions are being dropped and this is more of a session / connectivity issue than a performance issue.

Perhaps give HP-LeftHand support a call and see if they can help.