[GE users] Fair-share with suspension

m0zes adam.tygart at gmail.com
Tue Jul 21 15:11:39 BST 2009


I have a had a similar problem. Here is what I did to work around it:

I have a per user slot quota of roughly 65% of the cluster (on top of
the "fair-share" implementation). I have also restricted jobs that
last over 72 hours to 80% of the cluster. These two things have helped
me all but guarantee that there will be a few cores free for
"high-priority" jobs.

Hope that helps,

Adam

On Tue, Jul 21, 2009 at 03:05, jjh<jjh at 42quarks.com> wrote:
> Firstly, apologies if this has been asked before, I hadn't a search of
> the archives etc. but I couldn't find an specific answer to my query.
>
> We are using SGE 6.2 with fair scheduling. It works very nicely.
> Except when users have long-running jobs then we have problems because
> this can stop others getting work done. We currently have a
> high-priority queue which will suspend jobs on the normal queue to let
> its jobs run. But people end up scheduling lots of jobs on the high
> priority queue as well.
>
> Is there a way to implement fair scheduling, but also have some
> setting such that if the "fairness" is getting too unbalanced (i.e.
> someone's long running jobs have taken up all the slots) then suspend
> user's jobs to allow other users jobs to run (these jobs are all on
> the same queue)? The fair scheduling currently works to the extent
> that other users jobs are at the top of the queue, but they still
> aren't being run because they are waiting for free slots.
>
> Regards,
> Jonny
>
> --
> Jonathan J Hunt <jjh at 42quarks.com>
> Homepage: http://www.42quarks.com
> (Further contact details there)
> "Physics isn't the most important thing. Love is." Richard Feynman
>
> ------------------------------------------------------
> http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=208550
>
> To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].
>

------------------------------------------------------
http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=208621

To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].



More information about the gridengine-users mailing list