sort-of-solved WAS: Re: problem: __alloc_pages: 0-order allocation failed

[Erik Steffl <steffl@bigfoot.com>]

Sebastian Bober wrote:
> 
> Am Don, dem 22.November 2001, um 05:27:32 -0800, schrieb Erik Steffl:
> >
> >   usually it kills more programs, system services (cron) etc... I have
> > few question related to this problem:
> >
> >   - is it possible it's kernel problem? I haven't heard anything too
> > weird about 2.4.10
> 
> well, 2.4.10 is quite bad at deciding which process to kill. you should
> use 2.4.14 or even better 2.4.15pre9 (which seems to become 2.4.15).
> 
> >   - how can I find (post mortem) which program caused the problem?
> 
> i don't think there is a way (except, of course, heavy logging).
> 
> >   - how does kernel decide which programs to kill? is there any way to
> > influence it? I would rather have netscape (usually a memory hog) killed
> > than system services (like samba daemons, cron, nfs daemons etc.)
> 
> the kernel uses some heuristics (which have been sane only since
> 2.4.14). unfortunately you don't have a way to tweak the behaviour.
> usually the kernel chooses the process which uses the most memory (and
> won't free any in a low memeory situation).
> 
> i suppose it would be best to upgrade the kernel and then try to
> reproduce the problem. chances aren't that bad that it is way better.
> 
> >
> >   related incident: I have quite a few mp3 files and I noticed that when
> > I sort playlist in xmms by song name it eats all available memory (and
> > does the same after I restart it, I didn't even find any relevant
> > setting in config file but when I remove it problem goes away). I have't
> > tried it before so I don't know whether it's new problem or old one (I
> > haven't find anything similar on xmms or debian page so I filed a bug
> > report)

  update on the above problem, I think the info might be interesting for
some users:

  the memory eater was xmms (I filed a bug, it's on it's way to
upstream)

  I tracked it down using top -b, it prints out the formatted
information to stdout in a form that can be redirected into file
(instead of updating screen it prints each screen one after another, as
text). I saved this to file, waited till the problem occured again and
then checked the top info from time when the kernel started to kill
programs because of memory being all used.

  per advice above I also upgraded to kernel 2.4.14, the xmms problem is
still there (no wonder, it's xmms problem) but instead of killing cron,
fetchmail, X or other seemingly random program the kernel killed xmms
each time it got greedy with memory.

	erik