Memory issues

[Luc Monnens]

Dear all,

I have an L3000 running hp-ux 11.0.
It has 4Gb RAM and 4Gb of Device swap.

Lately my available memory has gone down from about 3400Mb to 50Mb when nothing is running at night.

I have no idea why an (almost) idle system needs all that memory.

Furthermore it leaves no memory to run production so it has to swap out like crazy during day time.

Any input on what this might be AND/OR on how to discover what is constantly keeping memory occupied will be much appreciated.


Kind regards,


Luc Monnens

[Dave Unverhau_1]

Luc,

The first thing I'd do is to execute the top command. Is there an obvious memory hog?

Next, I'd execute the dmesg utility (to see if all of memory is working at boot time) and check the syslog (/var/adm/syslog/syslog.log) to see if anything is happening that is causing memory to be deallocated.

Do you have online disgnostics installed? STM might provide some valuable info.

Do you have Predictive or its replacement (ISEE) installed? Are these logging any memory-related events?

How long has this been occuring? Have any software changes been made that might account for a memory leak? Any new patches added?

Do you have programmers developing code on this system? (Hopefully not on a production box, but I thought I'd mention it...)

Post what you find and somebody here should be able to help you zero in on the source of your grief.

Best Regards,

Dave

[Joseph Loo]

hi,

need to appreciate what applications or database running on that server. how will u know if it is idle when u have nothing to monitor in during the night.

a cron running simple script:

vmstat 5 300 > /tmp/memory

will help u check if the free memory really drops during the night.

regards.

[Luc Monnens]

Hi Dave and Joseph,

Thank you for replying.

I posted this on the hp-ux system administration forum as well - which I guess is a better place for this type of question.

To answer your remarks
Dave,
Nothing in top, dmesg or syslog.
I am not sure where STM can help. Can you tell me?
No Predictive or ISEE.
This started about 4 weeks ago and gradually became worse.
No new software or new patches.
No programmers writing code.

Joseph,
Idle means there is no database or application activity.
I'll try the vmstat tip.

I will give points to your answers, but I would prefer to continue with this thread on the other forum.
Thnx again.

Luc Monnens

[Bill Hassell]

Try ipcs -bmop and look for a very large shared memory segments, especially where NATTCH = 0 and KEY = 0. These segments were orphaned by processes that crashed or worse, by sysadmins that use kill -9. These unused segments can be cleaned up with ipcrm.

Check the kernel parameter dbc_max_pct and make sure it isn't 50. Change it to a value that works out to be about 400-600 megs.

Finally, summarize all the current processes by memory usage:

UNIX95= ps -e -o vsz,pid,ruser,args | sort -rn | more

This will show all the processes sorted by memory usage. You may have a rogue/runaway process that has grabbed thousands of megs.

[Luc Monnens]

Sorry to have not closed before.
This thread was moved to the sysadmin section - where I should have started it anyway.