On Fri, 24 May 2013, Ray Pete wrote:
Below was the patch.. which seemed to go somewhat (with a little work)
nicely into the source for me to build a srpm/rpm.
http://gridengine.org/pipermail/users/2012-September/004699.html
Oo oo! That was me!
This was the latest set of those patches (I moved it to the SoGE list to
avoid spamming this one with details irrelevant to the other forks):
http://arc.liv.ac.uk/pipermail/sge-discuss/2012-December/000343.html
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 explain why..
...
I had problems with interactions between the memory cgroup and Lustre
filesystems. I needed to get something into production, so switched to
improving how gridengine measured memory usage - see the ENABLE_USEDMEMORY
feature in the above post - and got rid of the virtual memory ulimit
(RLIMIT_AS), leaving the cgroup work to finish off in the future.
ENABLE_USEDMEMORY has been very successful locally, so I'm not sure if
I'll go back to looking at the memory cgroup.
I think Dave's done something similar in more recent versions of SoGE and
enabled it by default.
Long term, I think the OGS folk's strategy of using cgroups to reduce the
complexity of the execd is a good one. Unfortunately, I don't recall
hearing much from them lately - I hope things are going well for them.
All the best,
Mark
--
-----------------------------------------------------------------
Mark Dixon Email : [email protected]
HPC/Grid Systems Support Tel (int): 35429
Information Systems Services Tel (ext): +44(0)113 343 5429
University of Leeds, LS2 9JT, UK
-----------------------------------------------------------------
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users
