Server restarts by itself

[Helen French]

Hi,

My HP 9K (K210) running on version 10.20 restarted by itslef. No error messages in the syslog. The shutdownlog gives "Reboot after panic: kalloc: out of kernel virtual space".
The same happened 3 months back with the same message. The system working fine now. Any reasons or solutions?

appreciated ur help!

[Sanjay_6]

Hi Shiju,

Looks like you have to tune your kernel parameters.

Take a look at the threads below,

http://us-support2.external.hp.com/cki/bin/doc.pl/sid=782658e71b86af7d09/screen=ckiDisplayDocument?docId=200000056144612

http://us-support2.external.hp.com/cki/bin/doc.pl/sid=782658e71b86af7d09/screen=ckiDisplayDocument?docId=200000051483143

Hope this helps.

Regds

[Uday_S_Ankolekar]

Hi,

Read this thread for more info on kalloc
http://us-support.external.hp.com/cki/bin/doc.pl/sid=4d273468089af38ebf/screen=ckiDisplayDocument?docId=200000051483143

-USA..

[Sanjay_6]

Hi Shiju,

You may need some patches too,

http://us-support2.external.hp.com/cki/bin/doc.pl/sid=782658e71b86af7d09/screen=ckiDisplayDocument?docId=200000055052600

Hope this helps.

Regds

[Sridhar Bhaskarla]

Shiju,

Looks like you need this patch PHKL_16751. Check if it is not already there. The description of this patch contains

//start

The system message buffer would show many "sysmap: rmap
ovflo, lost [...]". Eventually, the system would panic with
"kalloc: out of kernel virtual space". This problem would
only be seen on PA8000 systems.


//end

If you dont' have this, try installing it.

-Sri

[A. Clay Stephenson]

Hi:

This was a rather common failure on 10.20 boxes if PHKL_16751 was not installed. It may have been superceded so check the patch database.

[Christopher McCray_1]

Sorry it took so long, but here are some instructions on doing a crash dump analysis(Q4). I would suggest doing this analysis and calling HP to assist with a course of action.

Hope this helps.

Chris

[Uday_S_Ankolekar]

Hi,

Some more info on patch
http://us-support.external.hp.com/cki/bin/doc.pl/sid=4d273468089af38ebf/screen=ckiDisplayDocument?docId=200000014941099

[Helen French]

hey,

thanks for the reply ...

i have the patch PHKL_16751 installed on the system. also, the nproc value is 901 !

so ...

[Don Morris_1]

If nproc is 901 and you're still running out of virtual space then raise nsysmap above 1802 - it's really impossible to say without real data how high you need it. Doubling is the usual rule of thumb, so try 3600.

Raising it will result in the kernel using more physical memory for it's internal
structures (I *think* it's 16 bytes per nsysmap element), so if physical memory is a bottleneck on your system, this would increase your likelihood of swapping.

I'd also try to monitor the logs and watch for the rmap overflow messages - try to see what's running at the time and immediately before that would be causing lots of small kernel allocations which aren't all freed. The overflows stem from kernel virtual address space fragmentation.

[melvyn burnard]

what is nfile set to?

If a high number, decrease it a little (25%)
If not, then try increasing nproc.
You will be able to see if you are heading for a problem as in syslog you will see:
sysmap:rmap ovflo

The more often this is logged, the sooner your system will probably panic.

Patching is one part of the story, the other two are the settings of nproc and nfile

HTH

[Uday_S_Ankolekar]

Hi,

What is number on sysmap...? Should nproc*2

-USA

[Helen French]

hey,

thanks again ..

nfile = 1851

and there is no 'nsysmap' parameter !

I have lots of 'sysmap: rmap ovflo' messages in the system.

so ...

[melvyn burnard]

ok, now you need to increase nproc (double it?) and then rebuild the kernel.
boot from th enew kernel and monitor syslog.log for the rmap ovflo messages.

[harry d brown jr]

When I saw the title "Server restarts by itself", I thought to myself that it would be a cool feature for MS products, but then again they do that already don't they?

sorry for digressing.

live free or die
harry

[Helen French]

if i increase the nproc value, will it affect the performance of the system?

the memory is always 80% used !

live free harry ...

[harry d brown jr]

Increasing nproc will allow more processes to run, and if that happens, then yes, you will have negative results. But if memory is only at 80% in use, why not try to use the rest?

If you know that the number of processes isn't going to go wacky on you, then increase it!
I say set it to 2048!

live free or die
harry

[Don Morris_1]

Hmm.. maybe nsysmap was introduced in 11.0 - I know it's there at least.

Well, to cover another possibility, is PHKL_16751 or one of it's predecessors on your box? It superceded PHKL_11165 which had a specific fix for fragmentation of the kernel virtual space due to superpage use. Best I could find in the patch catalog, and that's really all I can recommend if 10.20 doesn't have nsysmap.

Don

[Don Morris_1]

Sheesh... that's what I get for not re-reading when I come back to the thread. You have the patch already.

Well, on 10.20 - if nsysmap isn't there, and you have the patch I don't know what else you could do (certainly nothing in the kernel) other than try to watch for what exactly is fragmenting the address space.

Sorry bout the repeated patch suggestion,
Don

[Helen French]

Hi Melvyn / Harry,

It's again ...

I changed the nproc size to '1800' and tested. Now there is no 'sysmap: rmap ovflo,' messages in my syslog.

Does this mean the problem is resolved ? give your feedbacks ..cuz the points waiting for u !

Thanks a lot
Shiju