High Performance Computing Group

 View Only
  • 1.  Fairshare not quite going to plan

    Posted Tue August 11, 2020 09:58 AM
    Hi all,

    I'm new to LSF, so please bear with me.

    At the moment we have a User X running a lot of jobs on our compute cluster. Although there is supposed to be a fairshare policy in place, it seems that if you have a 'large job' i.e. 256 cores requested in the queue, User X jobs take precedence over them because his/her jobs are smaller, say 128 cores, and the job scheduler fills the nodes with any jobs that will run on available resources.

    I would like to strike a sensible balance here because it seems unfair.
    Are there any examples of how to resolve such a situation?

    ------------------------------
    Gerald Davies
    ------------------------------

    #SpectrumComputingGroup


  • 2.  RE: Fairshare not quite going to plan

    Posted Tue August 11, 2020 10:18 AM
    Consider to use plan based scheduling in this situation, so the scheduler will reserve free cores for 256-core job till it get all 256 cores for running.
    https://www.ibm.com/support/knowledgecenter/SSWRJV_10.1.0/lsf_admin/chap_plan_based_scheduling.html

    -Yi

    ------------------------------
    YI SUN
    ------------------------------