diagnosing OOM issue

Dave Mitchell davem at iabyn.com
Fri Apr 16 14:04:13 UTC 2010


On Fri, Apr 16, 2010 at 03:40:51PM +0200, Oliver Falk wrote:
> You should not run a Linux system without swap. If a process tries to allocate memory and no more real memory is available and also no swap OOM will start killing.
> It's a workstation I guess? 

yes

The point being that I'm *nowhere near* exhausting 2Gb of RAM, and the OOM
is still playing Russian Roulette with my system.

I removed swap for the following reason:

Originally the system had 1 Gb RAM, ran FC9 (IIRC) and everything was
fine. Then I installed  F11, and enabled disk encryption. After this,
whenever memory usage rose to about 512Mb (ie 50%), the system would start
swapping everything out like mad. This would usually completely lock my
system up up for a minute or more (cursor frozen etc). I played around
with the vm.swappines setting, first setting it to 20 and eventually 0%,
and this made no significant difference. So I bought more RAM, doubling it
from 1Gb to 2Gb.  The swap freezing continued, only now when usage reached
nearly 1Gb. So I turned off swap, and now the system will allow me to
allocate up to 2Gb without mad freezy swapping. *Except* that it also
decides that there there isn't any low memory left (or something) and
decides to kill things off instead.

So, Dear Linux Kernel: I have 2Gb of RAM on a 32-bit OS; how do I set
things up so that I can run processes that collectively run up 1.5Gb
(say) of VM usage without my system freezing or random processes being
killed?

(In all the above I'm taking of course about usage ignoring buffer/cache
usage, i.e. what the '-/+ buffers/cache' line of free reports).


-- 
In economics, the exam questions are the same every year.
They just change the answers.


More information about the users mailing list