Slower XP guest performance on vSphere

[bellym]

We have a bit of  aweird problem and although a call has bee raised with vmware I just wanted to see if anyone has seen similar.

We have an XP guest vm that has been running on ESX 3.5 u5 for some time now without problems.

Recently we upgraded the esx farm it was running on to esx 4.1 u1 and although it appears to be running normally certain operations take 5 times as long.

For example a user complained that a spreadsheet that used to take 20 seconds to open, now takes over 90 seconds, his colleagues produced several other spreadsheets doing the same. We then realised that opening any of these sheets on any XP guest running on vSphere experienced the same problem.

We decided to move the guest back to 3.5 and the performance returns to normal, so we then rebuilt one of the vSphere servers with 3.5 and again the performance was normal.

I've tried ESX 4.0, ESX 4.1, and ESXi 4.1 and all exhibit the slow down in performance, even with only a single VM running.

We moved the machine to local storage and this made no difference. ESXTOP does not display anything unusual.

Hardware is HP Proliant c7000 Blade chassis with BL680c G5 blades.

Any ideas?

[J1mbo]

What switches are the new servers running through?

[bellym]

They are plugged into Cisco switches but I'm pretty sure network is not involved either as I copied the file locally and removed the virtual nic from the VM and the problem still persists.

It's like it can't get enough of the CPU, even though esxtop shows no problems with CPU resources.

[admin · Accepted solution]

Is this a 32-bit guest?  It is possible that ESX 3.5 was using binary translation for this guest, but ESX 4.x is using hardware-assisted virtualization.  Try changing the preferred execution mode (under ESX 4.x) to binary translation.  See http://www.vmware.com/files/pdf/perf-vsphere-monitor_modes.pdf for instructions.

[bellym]

That fixed it.

I'd felt sure we had tried chaning the BT setting (well one of my colleagues said he tried it so stupidly I believed him).

Moral of the story - "make sure you see it with your own eyes".

Thanks for your help.

[admin]

Good to hear.  My guess would be that you are running some anti-virus software in the 32-bit Windows XP guest that induces a lot of TPR activity, which can degrade performance under hardware-assisted virtualization.  That would put you into Scenario 2 in the referenced paper.