[Gllug] Forcing CPU and Mem limits on a process

[Nix]

On Mon, 14 Feb 2005, Richard Huxton whispered secretively:
> Nix wrote:
>> Er, what results would you expect if you tried to actually use more
>> mem+swap than you had? I'm having trouble thinking of non-surprising
>> semantics here...
> 
> I think the problem is not so much that the kernel goes around killing
> processes to reclaim memory. Rather, AFAIK you can't tell which of
> your processes will die ahead of time. In particular, I believe a
> large database process can look very attractive as a target when
> reclaiming memory, whereas you'd probably want almost everything else
> killed first if possible.

The intention of the OOM killer was always to try to avoid killing
things like that (and things like the X server...)

But it's a very, very hard problem, so the OOM killer often guesses
wrong, and the penalty for guessing wrong is extreme...

-- 
Synapsids unite! You have nothing to lose but your eggshells!
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug