Performance & VMmark : Interpreting esxtop Statistics : Comments

RE: Interpreting esxtop Statistics

Tomas NL — Tue, 06 Jan 2009 08:45:07 GMT

Very good article! This wil help a lot of us!

Tomas
www.tendam.info

RE: Interpreting esxtop Statistics

trusted@dvisor — Tue, 13 Jan 2009 05:20:53 GMT

Thank you! Awesome Doc!

Sean Clark - http://seanclark.us

RE: Interpreting esxtop Statistics

Iwan Rahabok — Fri, 06 Feb 2009 09:08:32 GMT

Scott, can we add things like this into the manual? Or have the manual point to this VMTN Performance page?

Just adding my 2 cents:
Q: What's the difference of CCPU% and the console group stats?
A: CCPU% is measured by the COS. "console" group CPU stats is measured by VMKernel. The stats are related, but not the same.
CCPU measures what is running in the SC world.
console also measures things that VMKernel runs on behalf of the console.
example, in an idle ESX, if you do a VM cloning, you'd see the CCPU% to be low. But the "console" world will be high.

Q: Is it possible that %USED of a world is greater than 100%?
A: Yes, if the system service runs on a different PCPU for this world. It may happen when your VM has heavy I/O.
Also, if the VMKernel is doing lots of work on behalf of that VM.
esxtop assign that work to the %USED on the VM.
What sort of works? I'm not too sure, perhaps I/O such as FC, NFS, iSCSI read/write.

RE: Interpreting esxtop Statistics

udaykiran — Tue, 10 Feb 2009 14:05:37 GMT

Very insightful document about the performance counters in esxtop.

I was wondering if there are any counters using which I can analyze the low level hardware behaviour like Cache Miss/TLB Miss? Please let me know if there is any workaround in vmware using which I can get these counters.

RE: Interpreting esxtop Statistics

dmitrif — Wed, 12 Aug 2009 18:56:29 GMT

%WAIT - %IDLE - 100% * (NWLD - NVCPU) = %IO
98 - 0 - 100 * (4 - 2) = -122
???

RE: Interpreting esxtop Statistics

zpan — Wed, 12 Aug 2009 19:28:42 GMT

As already mentioned in the document, the above formula is a very rough estimate. For different ESX versions, the number of worlds for a VM may not be the same. (New versions may introduce or remove some supporting worlds.) I guess this might be the reason you got a negative number. You might want to expand (using command 'e') the group and look for the VCPU worlds, and compute (%WAIT - %IDLE). I have to emphasize again it is not really %IO. %WAIT includes lots of different events.