[gridengine users] cgroups integration with ogs 2011.11p1
rpete at broadinstitute.org
Thu May 23 23:39:02 UTC 2013
Below was the patch.. which seemed to go somewhat (with a little work)
nicely into the source for me to build a srpm/rpm.
I never actually used it in production. Just tested it to see where it was
and get a feel for how it would work out.
There was not a lot of traction on it after this.. and your points would
On Thu, May 23, 2013 at 7:14 PM, Ray Pete <rpete at broadinstitute.org> wrote:
> Hi Dave,
> Interesting indeed, I will have to review my rpm again. it's been quite a
> few months since I touched it, but maybe it was a patch you posted some
> time ago
> that I applied and built to test out. I can say the cgroup directory
> created in /cgroup/memory is the gridengine job id and seemed to work fine
> in my very simple malloc cases.
> What do you mean by performance effects? Did you throw this out in favor
> of better methods to manage host based resource starvation.
> I am most interested in why it was canned because I was quite excited to
> see it being developed?
> On Thu, May 23, 2013 at 6:55 PM, Dave Love <d.love at liverpool.ac.uk> wrote:
>> Ray Pete <rpete at broadinstitute.org> writes:
>> > Udo,
>> > *Assuming you are running a RH like distro and your service cgconfig is
>> > running*
>> > It was quite some time ago I set this up but If I remember I had trouble
>> > with 2011.11p1 for some reason. ( probably user config error on my part
>> > )
>> I don't think there is any such support in that.
>> > I ended up using SoG 8.1.1-1 with the "CGROUP_MEMORY=/cgroup/memory"
>> > The one snag I found was unless you mess with cgroups and its
>> > you have to run sgeexecd as root to allow access to the /cgroup/memory
>> > directory.
>> I'm confused; that also doesn't have cgroup support.
>> > submit a sleeper job with h_vmem resource request and you should see
>> > job directory created in /cgroup/memory with the memory restrictions.
>> > I didn't spend too much too to be honest, but the simple tests I ran
>> > it to be working as expected.
>> In what way, exactly?
>> > Hopefully, others on the group have had more experience with the setup,
>> > I would be keen to here other setups and possible solutions.
>> I haven't found time to finish off work on memory and other cgroups, but
>> the whole thing is a can of worms for various reasons. Note that using
>> the memory cgroup is alleged to have significant performance effects.
>> What are you wanting to use it for?
>> Community Grid Engine: http://arc.liv.ac.uk/SGE/
> Raymond Pete
> Scientific Computing Services
> 617-714- 8202
Scientific Computing Services
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the users