what kernel parameters to monitor on hpux servers

[John J Burke]

Hi all,

I am a junior admin, supporting 3 hpux11.0 servers. We are moving from candle to netiq monitoring, in the past we just used candle for swap, file systems, cpu load and virtual memory. We have had an issue last year where the nfile parameter reached its max. I would like to try and avoid a situation like that again. Can you suggest for me all kernel parameters that i should monitor, so nothing is missed this time. Any ideas on what should be monitored on a hpux server would be appreciated, or links to forums disscussing these issues.

Thanks

John

[Pete Randall · Accepted solution]

John,

I would suggest monitoring what Glance (HP's performance monitor) monitors:

nproc
nfile
shmmni
msgmni
semmni
nflocks
npty
nbuf
buffer cache
swap space


Pete

[Kent Ostby]

Add in "ninode" to that list as I've seen that table fill up and cause problems.

Also, check that your systems are configured with 2x RAM worth of device swap at least.

[Eric Antunes]

Hi John,

To monitor the refered parameters (add dbc_min_pct and dbc_max_pct), use "sar" command:

To monitor overflows (ninode)

#sar -v 3 333

To monitor buffer cache (dbc_min_pct and dbc_max_pct)

#sar -b 3 333

...

Best Regards,

Eric Antunes

[Arunvijai_4]

Hi John,

This should be helpful..

http://www.hp.com/products1/unix/webservers/apache/techtips/index.html

-Arun

[Victor BERRIDGE]

Hi john,
à t depends what is running on your servers...
To what you already received I would add the "per user" params:
maxfiles and maxfile_lim, maxuprc (high if java running)and max_thread_proc, but you know, once correctly tuned it is very rare you have to deal with these parameters, not that I dont believe in monitoring (very good thing) but more you by monitoring should be proactive and feel when a little tuning has to be done before problems happen and negociate downtime because there are still parameters that are static and you will have to reboot...


All the best
Victor

[Jeff Schussele]

Hi John,

As Victor points out, I would definitely add maxuprc to your list.
"Generic" users like oracle will definitely hit that wall well before nproc.

My 2 cents,
Jeff