Re: [debian-knoppix] Undefined behaviour after some uptime
Re,
On Fri, Jan 03, 2003 at 10:34:17PM +0100, Andreas Kostyrka wrote:
> Am Freitag, 3. Januar 2003 17:07 schrieb Valentijn Sessink:
> > At Fri, Jan 03, 2003 at 01:13:41PM +0100, Nils Magnus wrote:
> > > longer time or when running memory consuming applications like mozilla,
> > > ethereal with large dumps, ettercap or stuff like that. At times either
> > > those applications or the KDE components seem to crash.
> > >
> > > I would expect the kernel to deny allocation of new pages/processes if
> > > there is not enough memory left but rather not crashing arbitrarily
> > > applications. Has anyone else experienced this behavior? Personally, I
> >
> > No, but I did not run Knoppix for longer time. But what you are
> > experiencing seems like the standard "opportunistic" memory management:
> > Linux hands out more memory than there's actually there. When a certain
> > program starts using a memory page that is not actually there anymore,
> > there is no solution but to kill of a random program.
Thanks Valentijn, that info was most helpful, I was not aware of this
implementation and was most probably influenced by the romantic but yet
naive ideas my professor told me a couple of years ago (in turn
influenced by Tannenbaum and other "dreamers" ;), that there should
never be more (possibly dirty) entires in the page tables than there is
physical memory plus swapspace.
Did I understand it right, that "opportunistic memory allocation" and
/proc/sys/vm/overcommit_memory are more or less the same?
> > There are patches that make Linux behave more correct - but this will
> > probably restrict your Knoppix usage even more.
> Well, this is something, AFAIK from the myth collection. Linux did overcommit
> memory. Notice that the sentence is in the past tense. Current kernels do not
> do this anymore by default.
>
> andreas@andi-lap:~$ cat /proc/sys/vm/overcommit_memory
> 0
> andreas@andi-lap:~$ uname -a
> Linux andi-lap 2.4.20 #5 Mit Jan 1 18:55:35 CET 2003 i686 AMD Duron(tm)
> Processor AuthenticAMD GNU/Linux
>
> Seems that overcommiting is nowadays a runtime switchable option :)
The short description from Rik was quite helpful:
http://linux-mm.org/docs/oom-killer.shtml
However, the unresolved question remains then, why I encounter these
situations still with a 2.4.20?
BTW the very report with the mozilla not starting up was different
anyway, it crashed with an SIGILL, as you might have read in my
attachment.
Well there are still confusing things, but some clouds have cleared.
Thanks ;)
Regards,
Nils Magnus
Program-Chair LinuxTag 2003 Free Conference Program
LinuxTag 2003: Where .com meets .org - magnus@linuxtag.org
_______________________________________________
debian-knoppix mailing list
debian-knoppix@linuxtag.org
http://mailman.linuxtag.org/mailman/listinfo/debian-knoppix
Reply to: