admin
Immortal
Immortal

Performance Monitor reporting CPU at 100%

The ESX performance monitor keeps reporting 100% CPU usage when WIndows Task Manager just averages at 50 - 60% CPU usage.

Anyone have an explanation as to why this is??

Thanks in advance.

0 Kudos
4 Replies
jhanekom
Virtuoso
Virtuoso

What OS is running in the guest? This usually a symptom of running the wrong type of kernel in the guest (e.g. multiproc in stead of uniproc).

admin
Immortal
Immortal

We have converted a physical Windows 2000 Advanced Server with 4 CPU's and 8GB of memory to a virtual machine with the same specs (4 vCPU, 8GB) using VMware converter 3.0.3.89816. The initial host was a HP DL380 G5 with 2 QuadCore CPU's and 8 GB of memory, running ESX 3.5 update 1.

0 Kudos
jhanekom
Virtuoso
Virtuoso

How many physical CPU cores are available in the host you virtualised onto? If that's also a 2x dual-core box, you may see better results if you reduce the number of CPUs in the VM to 2.

It's usually a good idea to keep the number of vCPUs in a VM to as few as possible, and generally a bad idea to have more than half the number of vCPUs per VM than you have physical cores. (This is usually revealed by high %READY times in esxtop, that is, figures well above 5-10%.)

In other words, if you have a 4-core physical server, don't create VMs with more than 2 vCPUs.

Some reference material you may find useful: http://www.vmware.com/pdf/vi_performance_tuning.pdf

Adidas6
Contributor
Contributor

I wanted to chime in and say that I am currently seeing something similar in our environment.

Details: ESX 3.5 update 1 with VC 2.5 udpate 1.

Host is a HP DL385 G1. 2x2 2.2GHz.

Guest is a Windows 2000 Server, 1 CPU. Tools verions 43424. (We think this may be contributing as it is old)

My fellow VM Admin found an issue with 3.5 Update 1 DRS causing a VM CPU to spike to 100% after a VMotion. We have plans in the next few weeks to update our landscape to 3.5/2.5 Update 2. Believe this may help solve this issue as well.

I'll follow up if that resolves this issue. (Unless we sort it out before then)

0 Kudos