Sampling interval

[konradclapa]

Hi everyone.

I got a question.

In Virtual Center there is a performance tab, where the chart showing the CPU can be viewed.

I see that there are some statistics of the CPU utilization: Minimum, Maximum, Average.

It also shows that the refreshing interval is 20 sec.

Does any know how many saples are taken during that 20sec? (it seems that there are more then 2 samples because the average value is not the average of max and min.

Can it be changed?

Cheers,

Konrad

[Rob_Bohmann1]

The metrics are collected every 20 seconds for ESX Server 3 hosts,

The default view shows the past hour of perf data, samples taken every 20 seconds. So there are 3 samples taken every minute. The samples are then averaged (I think in 5 minute chunks so 15 samples are averaged to create the view that you see by default in your client. At least that is how I understood it back in vc2,0)

Here is the doc that explains it all ad nauseum..... starts in Chapter 18 p285

You can change the time frame to day week or month, with the law of averages making the data less meaningful generally speaking, You can change the level of data colllected (amount of detail 1-4)

as well as adjust the intervals, Happy Reading

[TomHowarth]

Thread move to the vCenter forum

Tom Howarth

VMware Communities User Moderator

Blog: www.planetvm.net

[konradclapa]

Hi Rob,

Many thanks for your answer

I guess this is truth what u are saying, but actually i am talking only about the real time performance charts that are displayed in Performance Tab.

example

so this chart refreshes every 20 sec.

So ok lets assume that the sampling rate is 20sec. Then what significations have

the columns Last , Maximum, Minimum, Average?

We can assume that Last is the value of the last sample because the value of the last is the one that is put on the plot.

But what is the meaning of maxiumum Minimum and average then? For me it seems like there are more then 1 sample taken in this 20 sec interval.

Any idea, beacuse i can not find anything about it in the forum or manuals the virtual Center

Cheers,

Konrad

[Rob_Bohmann1 · Accepted solution]

you're welcome.

My understanding of this is the max, min, and average values are the values for the entire host over the time period show in the display, in this case over the last hour.

If you highlight the value that shows the percentages for the entire host, (I think its purple in your screen shot) you should see that the values in the chart equal the values in text in the table below the chart.

So when it says real time display, it is not moment to moment real time, but a chart of a sample taken every 20 seconds and then displayed on the chart in 20 second increments over the last hour.

In other words every 20 seconds it samples the host and then displays that value to the far right (end of the chart). The latest value is just that, the latest sample measured.

The max, min, and average value are those respective measurements over the last hour for the chart you posted. You are absolutely correct in thinking that the last sample cannot have different values for all those different measurements at the same time. The only time they would be the same is on the inital sample.

Another interesting thing to notice is you can see how well the scheduler is utilizing and spreading the load over the various cpu's by looking at this chart. When you look at each of the varios indivdual cpu's, you can see that each of the 4 cpu's have very similar max, min, and average values over the last hour, with a slight exception of a higher average for cpu0. This is to be expected since this is the cpu where the service console runs on, and so there will be a somewhat higher load on this cpu, generally speaking,

12 vm's on 4 cpus averaging about 30% cpu utilization. Not bad, 3 vms per core. Are they all single processor vms?

[Rob_Bohmann1]

PS...

If you want real moment to moment perf data, or at least a faster sampling rate, you can use the esxtop command run as root in the service console. The default refresh rate there is 5 seconds and when you look at that data, you will see the utilization rates bounce around a lot more (much greater variability) then what you see generally in the virtual center.

They recently posted a terrific document on the esxtop commands here http://communities.vmware.com/docs/DOC-9279

[konradclapa]

Yeah you are right concerning those values. I got confused with it and i could not find any exact information about it in manuals.

But now it is clear.

Well concerning the more acurate data, I think that i don´t need that actually.

For my monitoring purposes it would be better to have the date form 5samples averaged durnig one minute and then refreshed showing the media utilization of the CPU.

Becasue now any pick in the plots is a triger for sending SNMP Trap.

But as far as i know there is no option for makeing this average or sending the Trap after 3 samples are above treshold.

Btw. This is not screen shot of my system posting of images did not work so i posted some screen from google :).

Many thanks

Konrad