VMware Cloud Community
farmand
Contributor
Contributor

Cpu wait vs cpu idle performance counters

Hello,

I'm running VMware ESXi, 5.1.0 on a machine with 24 logical processors (2 sockets, 6 cores each, hyperthreaded)

When I get a look at performance counters through PowerCLI interface, I get results which I don't fully understand.

Here's the result I get, from which I've removed data on individual processors.

PowerCLI> get-vmhost "aaa.bbb.ccc.ddd" | get-stat -Stat cpu.idle.summation,cpu.used.summation,cpu.wait.summation -MaxSamples 1 -Realtime

MetricId                Timestamp                          Value Unit     Instance

---------               ----------                         ----- ----     --------

cpu.idle.summation      05/02/2014 11:28:20               185574 milli...

cpu.wait.summation      05/02/2014 11:28:20              1024833 milli...

cpu.used.summation      05/02/2014 11:28:20                62830 milli...


If I sum idle and used times I get 185574 + 62830 : 248404 ms

As I have 24 logical processors I assume that the overall elapsed time should be 20 000 ms * 24 = 480 000 ms...

I know that there are other times not accounted in idle or used... Hence I got a look at the wait time... Which is much higher than the 480 000ms...


Hence:

  • Is the following assumption valid?  wait + idle should be equal to 20 000 ms x nb_of_logical_processors,
  • Were is the time in between Used + idle and 20 000 ms x nb_of_logical_processors, the ready time, swapwait were all negligible when I looked at them.


Thanks for any help/hint

François

Tags (3)
Reply
0 Kudos
2 Replies
vmMarkA
VMware Employee
VMware Employee

The math is not as simple as that.

Here are some good references to start with outlining the stats and what they represent:

Interpreting esxtop 4.1 Statistics

VMware vSphere 5.1

Technical Marketing, Performance Specialist, VCDX @vmMarkA, virtualizationeh.ca
Reply
0 Kudos
farmand
Contributor
Contributor

Thanks for the pointer on esxtop. The other one I already went through the pages on CPu performance counters many time.

Unfortunately, I'm afraid it either does not shed any light in my mind, and may raise even  more questions.

All I've been able to grab from esxtop is the discussion on PCPU  regarding %USED vs %UTIL. All of this is discussing %, not millisec. However I may (wrongly) guess that used time may be split over  the two hyper-threads of a PCPU... Which would lead (assuming frequency is fixed) to at most 20 000ms x 12 = 240000 ms... whereas I get 248 404...

And I did not find any source of explanation for the cpu wait value.

Could you please be kind enough to elaborate a bit more?

Thanks in advance for your time and help,

François

Reply
0 Kudos