VCenter performance vs. guest performance

[sphinx99]

So, I have a VM under ESXi and I'm just copying some files to it, to test it out.

According to the vsphere client, that VM is pegged on CPU on the two vCPUs it's been given on a quad-core physical box. (No other VMs running.) CPU usage = 100%. MHz usage = ~ 6,000MHz which is about two cores.

I log into the vm, do a mpstat, and it's basically 80% idle, 20% system. That's what I'd expect. It's not doing much.

So does the vsphere client simply report incorrectly? Who do I trust?

 

[AceXsmurF]

Do you have the vmware tools installed on the guest os?
What VM machine version is it?

 

[sphinx99]

The VM is Solaris 11. I do have the latest tools installed. Are the tools the mechanism by which vsphere collects performance metrics? I would have thought it would get them directly from the hypervisor.

 

[defuseme2k]

It is probably not the guest vCPUs, its probably one of the processes charged to the VM -- perhaps its the mks process. If you're using ESXi with tech support mode enabled either via SSH or on the console -- login and run esxtop. Look for that virtual machine in the list, type e to expand, and then type in the id of the VM and press enter. That will expand the CPU statistics and will show you the CPU utilization per process.

 

[NetJunkie]

Yep. What defuse said. Here is a doc on esxtop. It can be a beast:

http://communities.vmware.com/docs/DOC-9279

May be the guest OS timer.