[GE users] Qsub strange behaviours

reuti reuti at staff.uni-marburg.de
Tue Jul 27 11:22:51 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. ]


Am 27.07.2010 um 12:07 schrieb spow_:

> Hi everyone,
> I have been running simple and parallel jobs altogether to see what would happen, and did observe behaviours I wasn't expecting at all. It might well come from the version I am currently using (6 N1) but I can't test on 6.2u5 until HP installed it on our cluster.
> - Our users will sometimes have to chose their queue (to prevent having their jobs preempted), so I tried to submit a few just to verify it works with:
> qsub job.sh -q queue(.q)

the syntax needs to be revised:

qsub [ options ] [ command [ command_args ]]

i.e. options come fist, then the command, the options to the command/script last.

> Which doesn't send the job where it is supposed to. It picks a queue randomly.
> While reading the manpages, I found the -hard option, but it doesn't work either.
> - Also, I use subordinates. The problem is that only part of the queue get suspended ! (e.g. my queue sub1 runs across 2 hosts, and only the sub1 at host1 gets suspended, whereas it is very clear the whole queue should be suspended).

How is the subordination defined in the queue setup (`qconf -sq ...`)?

> - Eventually, I ran parallel jobs, with $round_robin allocation. If I submit a limited number of jobs, they get correctly dispatched.
> But if a few jobs are already running in the parallel queues,

Why do you have many parallel queue? The idea behind SGE is to specify resource requests, and SGE will select an appropriate queue for your. It's not like Torque, where you submit into a queue.

> SGE dispatches the next job on both the queues. This behaviour is really bad for our cluster, as one parallel job really shouldn't span across multiple queues.

The is intended: when one and the same PE is attached to different queues, you can get a mix of slots. But it can be changed by using wildcards in your PE request:

Define one PE e.g. mpi1 for queue1.q and a second mpi2 for queue2.q (only attached to one queue each) and submit with `qsub -pe "mpi*" 4 ...`. Once a PE is selected by SGE, it will only collect slots from the queues it is attached to.

-- Reuti

> Are these 'bugs' (for all I know it might be on purpose) ?
> Is the behaviour the same in 6.2u5 ?
> Thanks for having read,
> GQ
> Votre vie privée l'est-elle vraiment ? Internet Explorer 8 vous prot?ge gratuitement


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

More information about the gridengine-users mailing list