Subject: Re: out of memory errors
From: Christopher Swingley (cswingle@iarc.uaf.edu)
Date: Fri Jul 25 2003 - 14:28:03 AKDT
Greg,
* Greg Madden <pabi@gci.net> [2003-Jul-25 13:04 AKDT]:
> jul 25 11:22:08 clos kernel: Out of Memory: Killed process 2846
> (vmware).
> Jul 25 11:22:08 clos kernel: Out of Memory: Killed process 2844
> (vmware).
This is the kernel telling you that something you're running has used up
so much memory that it's having trouble operating (the OOM killer, I
think they call it on the kernel list). Rather than allow the computer
to grind to a halt, it starts killing the processes it thinks may be at
fault. The heuristic is pretty good, but not perfect.
On the systems I manage, matlab can often cause this sort of problem
because my users sometimes try to load much too much data into matlab for
processing. Once the kernel has trouble, it kills matlab (usually).
Sometimes it makes a mistake and starts by killing X, which in turn
kills everything else the user was running.
My guess is that something running in vmware tried to take too much
memory (Windows RPC buffer overflow vulnerability?) and that brought
Linux to it's knees. You might look into the 'limit' command (or
'ulimit') for the shell you and your users are using. With a properly
contrained system, this problem shouldn't reoccur.
If the system is acting strangely in other ways, you might suspect the
memory itself, but there's no reason from your kernel log alone to
suspect it. I've found 'memtest86' does a great job of identifying bad
RAM if you think this might be the cause (again, I'm not saying there's
evidence to suspect that).
It could also be bad blocks on your swap partition / files, so that's
something to look at if you're sure it's a hardware issue and not one
program trying to hog all the systems (healthy) resources.
Another reason to suspect vmware is that it loads modules into the
kernel, so it's got greater access / greater potential for harm than a
normal job.
Chris
-- Christopher S. Swingley email: cswingle@iarc.uaf.edu IARC -- Frontier Program Please use encryption. GPG key at: University of Alaska Fairbanks www.frontier.iarc.uaf.edu/~cswingle/--------- To unsubscribe, send email to <aklug-request@aklug.org> with 'unsubscribe' in the message body.
This archive was generated by hypermail 2a23 : Fri Jul 25 2003 - 14:24:58 AKDT