You are viewing an old version of this page. View the current version.

    Compare with Current View Page History

    « Previous Version 15 Next »

     Click here for details...

    Hosts
    Explicit list of Worker hostnames to run the job on (comma-separated). Click 'Browse' to choose from a list of workers that you want to run the job on.

    Groups
    Explicit list of Worker groups to run the job on (comma-separated). Groups identify machines through some attribute they have, eg, a GPU, an amount of memory, a license to run a particular application, etc. Jobs cannot migrate from one group to another.

    Omit Hosts
    Explicit list of Worker hostnames to not run the job on (comma-separated). Click 'Browse' to choose from a list of workers that you do not want to run the job on.

    Omit Groups
    Explicit list of Worker groups to not run the job on (comma-separated). Click browse to choose from a list of groups that you not want to run the job on.

    Priority Cluster
    Worker cluster that the job has elevated priority to run on. Click 'Browse' to choose from a list of Priority Clusters.

    Clusters are a way of distributing jobs across the farm with the ability to expand and shrink based on the submitted priority cluster. Jobs can overflow from one cluster to another.

    Example:

    • A job submitted to /root/ will run on all machines with /root/ assigned as a cluster at a standard priority.

    • A job submitted to /root/a will run on all machines with /root/a assigned as a cluster at a higher priority than /root/ meaning that the job will pre-empt a job without the priority cluster.

     

    Host Order
    Order to select Workers for running the job (comma-separated) [+ means ascending, - means descending]. Click 'Browse' to choose from a list of Host Order Options.

    Host Order is a way of telling the job how to select/order workers

      • "+host.processors.avail" means prefer workers which have more slots available

      • "+host.memory.avail" means prefer workers which have more memory available

      • "+host.memory.total" means prefer workers which have more total memory

      • "+host.processor_speed" means prefer workers with higher cpu speeds

      • "+host.cpus" means prefer workers with higher total cpu slots

     

    Requirements
    Worker requirements needed to be met for job to run on that Worker (comma-separated, expression-based). Click 'Browse' to choose from a list of Host Order Options.

    Requirements is a way to tell the workers that this job needs specific requirements to be met in order to run. The drop-down menu allows a choice of OS:

    • "winnt" will fill the field with "host.os=winnt" which means only run on Windows based workers

    • "linux" will fill the field with "host.os=linux" which means only run on Linux based workers

    • "osx" will fill the field with "host.os=osx" which means only run on OSX based workers

    Some other options you can type directly: (is this all of them? )

    • "host.processors.avail.=4" means only run this job on workers that have 4 or more slots available

    • "host.processors.used=0" means only run this job on workers with 0 slots in use

    • "host.memory.avail=400" means only run this job on workers that have 400 memory available

     

    Reservations
    Worker resources to reserve when running job (comma-separated, expression-based). Click 'Browse' to choose from a list of Reservation Options.

    Reservations is a way to tell the workers that this job will reserve the specific resources for this job.

    Menu items:

    • "host.processors" this will fill the field with "host.processors=X" which means reserve X slots on the worker while running this job

    • "host.memory" this will fill the field with "host.memory=X" which means only reserve X memory on the worker while running this job

    Other options:

     

    Restrictions
    Restrict job to run only on specified clusters ("||"-separated) [+ means all below, * means at that level]. Click 'Browse' to choose from a list of Restrictions Options.

    Restrictions is a way to tell the workers that this job can only run on specific clusters

    Examples:

    • submit a job that will run only in /farm:

      "qbsub -cluster /root -restriction /root"

       

    • submit a job that will prefer to run in /farm/nuke, but could run in any host in /farm or in the first level below /farm

      "qbsub -cluster /root/a -restriction '/root or /root/*'"


    • submit a job that will prefer to run in /farm/nuke/workstations, but could run in any host at any level at /farm or below

      "qbsub -cluster /root/a/workstations -restriction '/root or /root/+' hostname"

     

    See Also
    • No labels