cpu ready question

[RUG201110141]

I know that high cpu ready may indicated poor performance of a virtual machine. In esxtop this statistic is given in a percentage. For instance a virtual machine may be in a cpu ready state 5% of the time. How does this correlate to the VirtualCenter cpu ready. It's given in milliseconds in the VirtualCenter performance charts.

[hicksj]

I remember discussing this with some folks a while back...

Took a few minutes to find it... in the archives:

http://www.vmware.com/community/message.jspa?messageID=462180#462180

[RUG201110141]

I'm not sure I exactly follow how the 5% cpu ready time correlates to 1000ms. I can see the math and how it's worked out in the posting but don't see the relationship. I have a daul proc vm that shows it's running around 10% wiith esxtop, but in VirtualCenter it's at about 900 to 1000ms.

[conyards]

5% CPU correlates to 1000ms (or 1 second) because the default sample rate within virtual centre is every 20 seconds.

5% of 20 is 1 second, therefore 1000ms.

As for the difference within ESXTOP and VC I can only speculate that you are looking at different samples.

How is the performance inside the server itself.

Simon

[hicksj]

esxtop updates every 5 seconds, while VC polls every 20 seconds. Is your %ready a constant 10% in esxtop? If so, I'm not sure why there would be an inconsistancy. Of course, my original theory could also be off

[RUG201110141]

Thank you for your help. I'm just not getting it though. I believe what you are saying but, how does a sample rate correlate to CPU usage. So virtualcenter refreshes the statistics of what is going on every 20 seconds. How is there a relationship between a sample rate and the business of virtual machines.

[conyards · Accepted solution]

The sample rate, is the frequency with which VC or ESXtop will check to see what CPU resource is being utilised. Hence the need to reference the sample rate when calculating CPU %ready time.

hence,

5% of 20sec is = 1 sec or 1000 ms

10% of 5sec is = 0.5sec or 500ms

1000 ms of ready time as reported by VC would be reported as 20% %ready time within ESXtop.

Hopefully this is a little clearer.

Simon

[jamesl73]

I still don't understand why you even need to consider the sampling time. Aren't the samples just a single point in time at the given interval? So whether you take the sample every 5 seconds or every 20 seconds the data means the same and shouldn't require different calculations.

If the samples were a sum of last 5 or 20 seconds then your calculations would make sense to me.