find cpu usage for the past event

[Fauziah Mahdan]

Hi there,
Is there a way I can get info from past event whereby the cpu threshold is high?
If can what and where is the log to look into?

Thanks
Fauziah Mahdan

[siva0123]

Hi ,
If you have monitoring tools like HP Openview or others then i hope you can have the CPU utilisation graph over a period of timeglance

I'm not sure Whether glance have a tack of the CPU utilisation.Any how u can give a try.

Thanks,
Siva

[Peter Godron]

Fauziah,
the answers will be the same as for your earlier thread, which you have not updated with your feedback.
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=1074326

The general principleis that past stats on usage etc. are only available if you have had some capture tool running at the time.

[Fauziah Mahdan]

Hi Siva,
I do habe openview operation. I found the cpu spike from this system. But I do not know what is the cause because it already past.
From the monitor alert it onl mention the percentage of the threshold. Never give the reason why it high.

That is why I ask how to trace the cpu spike for the apst event?
I open another thread from openview how I can create autocommand to run top whenever cpu spike.
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=1074350

[Yogeeraj_1]

hi Fauziah,

I use measureware to monitor/extract performance information about past events...


what you would typically need:
# swlist -l product|grep -i measureware
MeasureWare C.03.58.00 MeasureWare Software/UX
#

The above installed daemon will collect the data for the whole day than you can just run the command to extract the data in the desired format. e.g.
/opt/perf/bin/extract -xp -r /home/yogeeraj/zgbltemplate -g -b today 00:00 -e today 23:59 -f stdout | /bin/mailx -s "MYSERVER: Measureware Performance figures " myemailaddr@mydomain.mu >/dev/null 2>&1


The delimited data file receieved can be used to plot a graph in microsoft excel.

hope this helps!
kind regards
yogeeraj

[Fauziah Mahdan]

I have measureware version C.03.35.00
Can you give the exact command to capture the info for date Nov 13 and append it into readable file. I don't have emails etting into all my unix servers.

[Michael Steele_2]

CPU spikes, or bottlenecks, source back to higher then usual number of processes, multiple jobs in the run queue, as well as 0% CPU idle time.

You can easily insert a 15 minute cron redirecting and saving the output for a week at a time. ( sar, ps, wc ).

Recently, a cpu bottleneck was discovered on one of my servers. It was related back to a high number of process threads all attached to the same oracle shared memory segments via the oracle library cache. ( ipcs, and note NATTCH column. )

Unfortunately, increasing the oracle libray cache didn't relieve the cpu bottleneck. It really came down to the number of processes running on the box. So it kinda comes down to, add CPU's and memory, or decrease the number of processes on the box, which can also mean reduce the number of users, or, job schuduling during off hours / after business hours.

[Fauziah Mahdan]

Michael, I also suspect the spike is coming from the application site (oracle process). But I need to get the black and white.
Scheduling the task for every some period will require some maintenance job for the log growth and also slightly will impact the server performance. That is why I am planning to add the auto execute command run when there is cpu spike at hp open view operation level.
But the auto command seems doesn't work.