[GE users] Making Very Important job to run

seandavi seandavi at gmail.com
Tue Jul 13 14:27:53 BST 2010

    [ The following text is in the "utf-8" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some characters may be displayed incorrectly. ]

On Tue, Jul 13, 2010 at 9:20 AM, spow <guillaume.quere at fr.thalesgroup.com<mailto:guillaume.quere at fr.thalesgroup.com>> wrote:
I'm a newbie so please don't yell if I'm wrong, but what is stated here
is IMHO false :

seandavi a écrit :
> Look into subordinate queues.  We have two queues, one subordinate to
> the other.  All users run in the subordinate queue and when we need to
> run a job immediately, we submit it to the other, non-subordinate
> queue.  Approximately the same thing can be accomplished by having one
> queue run with nice 10 or something like that and a second queue run
> with normal nice level.
Nice values are working the other way : -10 gives a higher priority than
0 which is also higher than 10. I don't know what way you meant it,
pointing that just in case.

You can set the "nice" level on a per-queue basis.

Also, nice values is a UNIX program that gives/changes priority in terms
of CPU. If the job is pending, it doesn't solve the problem.
As for the subordinates I also find it very useful to not get into
mind-wrapping configurations, though it would mean that users have to
select a queue each time they submit a job (except if it's different
jobs types)

Yes, you would need to have users submit to different queues or have another way for SGE to select a queue based on consumables or some such thing.  In our case, we have all jobs associated with a certain project going to the non-subordinate queue.  As long as the non-subordinate queue is not fully subscribed, then these jobs run immediately and the subordinate queue suspends.

Subordinate queues are difficult to think about, I fully agree, but I think that is the best way to go in this situation.


More information about the gridengine-users mailing list