Re: Terminate process that fails on a constrained allocation

[Paul Jackson <pj@sgi.com>]

Can anyone give us a good reason why we shouldn't just remove the oom
killer, entirely?

Christoph wrote:
> If a task has restricted its memory allocation to one node and does 
> excessive allocations then that process needs to die not other processes 
> that are harmlessly running on the node and that may not be allocating 
> memory at the time.

That _exact_ same argument applies to a system that only has one node.

If we want to remove the oom killer, lets just remove the oom killer.


> People are accustomed of having random processes killed? <shudder>

That's what the oom killer does ... well, it makes an honest effort
not to be random.

So, yes, since it has been there a long time, people are used to
it.  Maybe they don't like it, maybe with good reason.  But it
is there.


> OOM killing makes 
> sense for global allocations if the system is really tight on memory and 
> survival is the main goal

If that argument justifies OOM killing on a simple UMA system, then
surely, for -some- critical tasks, it justifies it on a big NUMA system.

Either OOM is useful in some cases or it is not.


-- 
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <pj@sgi.com> 1.925.600.0401