High CPU when High Disk I/O

[yougotiger0]

I have several Windows 2008 R2 servers running in our vSphere  environment. We're seeing issues that during heavy disk operations (for instance, copying a folder with lots of files). That the CPU spikes often pegging out at 100% for the duration of the operation. I believe this may be happening on our 2003 servers as well, but haven't done any real testing.

Here's my question. I don't believe we should be seeing this spike in CPU should be happening, and it's causing difficulty with our email system. Can anyone recommend anything I should do to determine wherein lies the issue and to resolve it?

We're currently running ESXi 4.1 with the latest patches that have been pushed down through the update manager.

All the VMware Tools are up to date.

We're running on Dell PowerEdge 1950 servers.

Our SAN's are Dell Equallogic 5500s and we're using iSCSI.

[AndreTheGiant]

Can you check the storage performance monitor to see if you have high latency?-

Install SanHQ that can help you to monitor if the problem is on the storage side.

Otherwise can be in guest memory by having a high swapping (in pagefiles), but this is simple to verify (check your memory setting).

I suppose that you already have a recent Eql firmware and Jumbo frame set correcty.

Andre

[yougotiger0]

@Andre,

We've got recent, but not the latest firmware, and I'll get with our SAN guy and see if we can check out storage performance. He's also our Network guy so I'll have to have him verify jumbo frames, but he believes we do have jumbo configured and running.

As far as guest memory, we're running about 45% in use, the rest available, task manager shows pretty consistently the following:

kernal memory

117MB Paged and 113MB non paged

and for physical memory:

Total........4095

Cached....2181

Available..2194

Free........17

[yougotiger0]

After talking to support, it seems that we needed to add another CPU to the server. The I/O queue wasn't getting emptied fast enough becuase the CPU couldn't handle the load. Adding a 2nd CPU seemed to make a significant difference. Luckily we were running Windows 2008 which doesn't have any HAL side effects from putting in a 2nd CPU, just shut it down, put in a 2nd CPU and then started it back up.