memory, CPU monitor

[Nguyen Anh Tien]

My server (RP 5470) is alway overload. Memory and CPU is alway 100% (by using glance). but when i gather data by using sar I found that swapin and swapout is zero at all day. I think some kernel parameter is not exactly true. I want to tune it but i have no doc that said about it.
glance is impossible to store data. is there any tool that allow me to store performance data ?
Anyone can help me.

[Patrick Wallek]

The Openview Performance Agent is what you want. Check out this site:

http://openview.hp.com/products/ovperf/index.html

[Sridhar Bhaskarla · Accepted solution]

Hi,

%usr at 90% %sys < 10 and %wio < 10 is quite healthy. It means the system is really being used unless you see any rogue processes in your 'glance' window. I would look at the top 10 processes and see what they are..

You can use glance in adviser_only mode to store the data.

-Sri

[Nguyen Anh Tien]

do you have document that show me how to run glance in adviser_only mode?. Where URL can I get it ?

[Patrick Wallek]

Advisor mode is mentioned in the glance man pages.

# man glance

For example:

# glance -advisor_only -j 15 > glance_advisor.out

The '-j time_in_seconds' option tells glance how often to update.

[Sridhar Bhaskarla]

Hi,

Try the example.

Put the following lines in a file called glance.conf

PRINT " CPU SYSTEM MODE UTIL -",gbl_cpu_sys_mode_util

PRINT " CPU USER MODE UTIL - ",gbl_cpu_sys_mode_util

PRINT " CPU TOTAL UTIL - ",gbl_cpu_total_util

RUn the command

glance -adviser_only -syntax glance.conf -iterations 2 -j 2

Should print the stats for CPU system mode, user mode and total utilizations. You can add as many metrics you want. Look at /opt/perf/paperdocs for more metrics.

You can put the above in cron to generate metrics.

Else consider using 'sar'. man sadc, sa1 and sa2. You can also get sarcheck from www.sarcheck.com for a minimal cost I believe.

-Sri