Open MPI logo

Hardware Locality Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Hardware Locality Development mailing list

Subject: Re: [hwloc-devel] Cgroup resource limits
From: Christopher Samuel (samuel_at_[hidden])
Date: 2012-11-04 22:28:21

Hash: SHA1

On 03/11/12 09:05, Ralph Castain wrote:

> System resource managers don't usually provide this capability, so
> we will do it at the ORTE level.

I would argue that the resource managers *should* be doing it - however,
I will also argue that the resource managers should be doing it via
hwloc (so I'm afraid it's not an out for you folks :-) ).

It's also worth remembering that the memcg code has an appalling
reputation with the kernel developers in terms of performance overhead,
for instance at the recent Kernel Summit numbers were reported showing a
substantial impact for just having the code present, but not used.

Following that a patch set was sent out trying to avoid that impact if
it's not in use which doesn't help here but does give a measure of the
performance hit:

# So as one can see, the difference between base and nomemcg in terms
# of both system time and elapsed time is quite drastic, and consistent
# with the figures shown by Mel Gorman in the Kernel summit. This is a
# ~7 % drop in performance, just by having memcg enabled. memcg
# functions appear heavily in the profiles, even if all tasks lives in
# the root memcg.

- --
 Christopher Samuel Senior Systems Administrator
 VLSCI - Victorian Life Sciences Computation Initiative
 Email: samuel_at_[hidden] Phone: +61 (0)3 903 55545

Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Mozilla -