Re: 2.6.8.1+patches: Still a memory leak with cdrecord

[Karl Vogel <karl.vogel@seagha.com>]

Frank Steiner <fsteiner-mail@bio.ifi.lmu.de> wrote in
news:412F27C1.6030100@bio.ifi.lmu.de: 

> I'm not able to really track the problem down, because it is not
> permanent. PCs that were leaking yesterday, went fine this morning
> for the first 2-3 CDs, then started leaking again. Another PC
> that was leaking first, is now burning CDs without problems. All
> that with DMA always off for all the burners. I've also one
> PC that is always burning CDs without any leak (and they
> are all running the same kernel and have the same distribution/
> packet selection installed...).
> 
> I've no idea what's causing that flip-flop between leaking/non-leaking
> behaviour, but some bug is still there. What can I do to provide
> useful debugging information?
> 
> Following is the output of the oom killer.

I'm not sure, but this sounds a bit similar to a problem I am seeing. Are 
you by any chance using the CFQ scheduler?! (elevator=cfq) If so, give 
elevator=as or elevator=deadline a go.

The problem I'm seeing is that processes that allocate large amounts of 
memory, are OOM killed or cause swap storms.

-- http://thread.gmane.org/gmane.linux.kernel/228156