[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [HTCondor-users] Paralell Jobs



Hi Marcos,

I suggest set the machine rank to 0.
To disable preemption by machine RANK.

And please check the following expression on the master node:

condor_config_val -negotiator PREEMPTION_REQUIREMENTS

Cheers,

Imre



On 2013.09.02. 14:58, Marcos M. wrote:
Thanks for your reply.
Could you give me the exact steps/commands to achieve this? I did the following test an didn't work:

all users had priority less than 14.
condor_userprio -setprio DedicatedScheduler@***** 15
condor_submit mpi_script

and NegotiatorLog showed this
"Preempting user@local (user prio=3.32, startd rank=0.00) on slot2@xxxxxxxxxxxxxxxxx for DedicatedScheduler@******* (user prio=14.99, startd rank=1.00)"

Perhaps is rank misconfigured?



2013/9/1 Imre Szeberenyi <szebi@xxxxxxxxxx <mailto:szebi@xxxxxxxxxx>>

    Hi Marcos,

    I have the same problem, and I found a dirty solution:
    I set the priority factor of the Dedicated scheduler higher than
    the serial jobs has.
    The negotiator choose first the job with
    higher priority (lower number). If the dedicated scheduler
    has the highest priority the parallel jobs will be started first.
    Unfortunately the dedicated scheduler does not care about the
    user priority it uses the job priority instead (as I understood the
    scheduling).

    So I suggest you, to set the priority factpr of the dedicated
    scheduler
    between 2 and 100 using the condor_userprio command.

    I also modified the PREEMPTION_REQUIREMENTS expression in the
    condor_config file, because I want to preempt only the standard jobs.

    Now it is better, but is is far from fair scheduling. So I plan to
    write a
    daemon which will modify the user priority of the dedicated scheduler
    in dynamical way and also modify the job priority of the waiting
    parallel jobs
    in the queue. A know it is more dirty solution.

    Cheers,

    Imre



    On 2013.08.30. 18:58, Marcos M. wrote:

        Hi i'm running condor version 7.6.6 and configured condor to
        run both serial and mpi jobs. The problem is that mpi jobs
        inmediately preemt running serial jobs no matter their priorities.
        I know this is kind of normal behaiviour, i'd like to konw how
        can I configure paralell jobs to start with a priority of for
        exaple 2.00 so it has to wait at least some time to preempt a job.

        I have one central submit machine with dedicated scheduler and
        the rest are mpi enabled execute machines.
        RANK is set to "Scheduler =?= $(DedicatedScheduler)"

        Any help is appreciated.



        _______________________________________________
        HTCondor-users mailing list
        To unsubscribe, send a message to
        htcondor-users-request@xxxxxxxxxxx
        <mailto:htcondor-users-request@xxxxxxxxxxx> with a
        subject: Unsubscribe
        You can also unsubscribe by visiting
        https://lists.cs.wisc.edu/mailman/listinfo/htcondor-users

        The archives can be found at:
        https://lists.cs.wisc.edu/archive/htcondor-users/


    _______________________________________________
    HTCondor-users mailing list
    To unsubscribe, send a message to
    htcondor-users-request@xxxxxxxxxxx
    <mailto:htcondor-users-request@xxxxxxxxxxx> with a
    subject: Unsubscribe
    You can also unsubscribe by visiting
    https://lists.cs.wisc.edu/mailman/listinfo/htcondor-users

    The archives can be found at:
    https://lists.cs.wisc.edu/archive/htcondor-users/




_______________________________________________
HTCondor-users mailing list
To unsubscribe, send a message to htcondor-users-request@xxxxxxxxxxx with a
subject: Unsubscribe
You can also unsubscribe by visiting
https://lists.cs.wisc.edu/mailman/listinfo/htcondor-users

The archives can be found at:
https://lists.cs.wisc.edu/archive/htcondor-users/