Re: 2.4.23-pre VM regression?

[Stephan von Krawczynski <skraw@ithnet.com>]

On Mon, 20 Oct 2003 00:21:55 +0100 (BST)
Ken Moffat <ken@kenmoffat.uklinux.net> wrote:

> On Thu, 16 Oct 2003, Andrea Arcangeli wrote:
> 
> >
> > sure. I think I already explained there are downsides in disabling the
> > oom killer for desktops where the offender task is normally the biggest
> > one too, but those downsides aren't something I care about given the
> > cases it gets right w/o it (i.e. huge-shm-SGA/mlock/oomdeadlocks). the
> > oom killer can do the wrong decision too sometime, and more
> > systematically as well.
> >
> 
>  Any chance of getting the oom killer back as an option ?  On
> 2.4.23-pre7 I made the mistake of trying to print a high-resolution
> photo (300ppi, A4 size) using ghostscript while I was in X.  I've only
> got 128MB memory and about 256MB swap on that box, which obviously
> wasn't enough (gs typically uses up to 300MB for a 200ppi A4 picture).
> Only problem was that X got killed instead of gs, which left the box
> unuseable.  Last time I saw the oom killer in action it actually saved
> me from having to reboot.

After having read numerous arcticles regarding oom
killer/conditions/bugs/features I believe the lack of the oom killer is
possibly the best scenario _without_ any user configuration. If you really want
an oom killer it seems obvious that what you really want is a _configurable_
oom killer. The simple approach in configuration is handing it a list of
processes it should _not_ kill. This means of course you need a user-space tool
to mark processes as not oom-killable, possibly implemented as a kind of
priority. Not set means "kill-at-will", set means top-priority gets killed as
last.
Did we already have this kind of proposal? :-)
-- 
Regards,
Stephan