average of total process CPU util is less than global average CPU util?

[sms_3]

I can view the global summary CPU situation and get the average value. However, when I use the extract command to get the process information and do (sum proc_cpu_total_util)/processsors/sample number, the value is less than first value. Why? some mistake?

[Duncan Edmonstone · Accepted solution]

By global cpu do you mean the metric gbl_cpu_total_util ?

Cos that's never going to be equal to sum proc_cpu_total_util)/processsors/sample

surely you should be looking at bycpu_cpu_total util, and *not* proc_cpu_total_util which is only going to show you CPU utilisation for a given set of processes that was "of interest" to scopeux (what's of interest being set in /var/opt/perf/parm). By default I'm pretty sure any process using < 10% of a CPU, or less than a ceratin amount of memory/IO is going to get ignored.
HTH

Duncan

[sms_3]

Thanks for the answer.
However, I wonder if the gap between gbl_cpu_total_util and sum/average proc_cpu_total_util is so large.

In my system, interested CPU <5%, mem <900MB
sample interval:1 min(1440/1 day), CPU num=4

I got 1 day samples of proc_cpu_total_util and calculated:
(sum all proc_cpu_total_util)/1440/4=12.7(%)
and the gbl_cpu_total_util=28.1(%)

Will the gap be so noticeable?

[Duncan Edmonstone]

sure why not? you'd only need 8 processes trundling along at an average 4% CPU utilisation across your sample to account for this. They'd get wrapped up in your gbl_ number but not in your proc_ number. I wouldn't be surprised if there are plenty of system daemons and management agent utilisties that fall into that bracket.

HTH

Duncan

[Emil Velez]

if you are summarizing the cpu utilization of all of the processes remember that measureware only logs the interesting processes so the processes logged are not the only processes that used cpu during that 1 minute interval.

[sms_3]

I made a mistake. The setting is "threshold cpu = 5.0, memory = 900, disk = 5.0, nonew, nokilled".

If the gap is so large, that means there may be many daemon or system processes occupying the CPU resource.

Maybe I need to decrease the CPU threshold 5.0 to get more data or I should change my question to how to get the interested process's average CPU utilization.

Thanks for your explanation.