Load average in hpux 11.0

[vaman]

load average is shooting up inbetween over 15, but not constant. now it is 0.39, how to track this

[Paul Torp]

there is an option to "top" that print the sides to a file in intervall you specify.

do a man on top.

-paul

[Chris Vail]

There are dozens of ways to do this. I like to use sar (man sar). Another common utility is top. The best utility is glance, but you have to pay for this. I usually use sar and extract the data to a comma delimited spreadsheet, then open it up in a spreadsheet program for analysis. In the past, I've done the same with vmstat. HP recommends glance (of course).


Chris

[Steven E. Protter]

top or glance or sar scripts will top this.

Load factor is not necessarily a problem. It is calculated by how may processes are awaiting cpu time.

If you have a lot of processes going on but system performance is normal, load factor is not a reason to start tweaking.

I recall a lecture by Bill Hassell at HP-World where he had a system with a load factor of well over 50 but that alone was not a problem.

If your system is providing poor response or has symptoms, then its time to collect data and act.

I'm attaching some sar scripts for your convenience.

SEP

[Eric Antunes]

Hi,

If the top process it terms of %WCPU or %CPU is a Oracle one, you can see more info runnng the following query:

select *
from v$session
where process = '';

Kindest Regards,

Eric Antunes

[Zeev Schultz]

You can obviously download Glance and use it
for evaluation period (60 days I think) here:
http://www.openview.hp.com/products/gplus/tc_gplus_0001.html

Good software allways costs money but it does
the job :)

[Peter Gulotta]

The best tool to track any performance metric is the Measureware Agent. It is the companion to Glance that actually logs all the metrics. Glance, top, sar, etc. are all ad hoc utilities. The sadc tool could be used, but it does not have nearly the range of the MWA. All that being said, the load average, or global run queue, prior to 11.11 includes processes that are on a short disk IO wait. So, if you have reports running, or your application routinely does small disk IOs, it would not be unusual to see this metric spike up and down.

Keep in mind that the global run queue (GBL_RUN_QUEUE) is the same as the 1 minute load average reported by top and friends, but that is not the same as the global priority queue (GBL_PRI_QUEUE) which tracks the number or processes waiting on priority.