<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
  </head>
  <body>
    <p>It shouldn't impact running jobs, all it should really do is
      impact pending jobs as it will order them by their relative
      priority scores.</p>
    <p>-Paul Edmon-<br>
    </p>
    <div class="moz-cite-prefix">On 4/30/2021 12:39 PM, Walsh, Kevin
      wrote:<br>
    </div>
    <blockquote type="cite"
cite="mid:CAE7EdicioWaa0AwbDAfcVotQi7cP8RbjNn_FkDgjRj+qQaiJ9w@mail.gmail.com">
      <meta http-equiv="content-type" content="text/html; charset=UTF-8">
      <div dir="ltr">Hello everyone,
        <div><br>
        </div>
        <div>We wish to deploy "fair share" scheduling configuration and
          would like to inquire if we should be aware of effects this
          might have on jobs already running or already queued when the
          config is changed. </div>
        <div><br>
        </div>
        <div>The proposed changes are from the example at  <a
href="https://slurm.schedmd.com/archive/slurm-18.08.9/priority_multifactor.html#config"
            moz-do-not-send="true">https://slurm.schedmd.com/archive/slurm-18.08.9/priority_multifactor.html#config</a>
          :</div>
        <blockquote class="gmail_quote" style="margin:0px 0px 0px
          0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">#
          Activate the Multi-factor Job Priority Plugin with decay<br>
          PriorityType=priority/multifactor<br>
          # 2 week half-life<br>
          PriorityDecayHalfLife=14-0<br>
          # The larger the job, the greater its job size priority.<br>
          PriorityFavorSmall=NO<br>
          # The job's age factor reaches 1.0 after waiting in the<br>
          # queue for 2 weeks.<br>
          PriorityMaxAge=14-0<br>
          # This next group determines the weighting of each of the<br>
          # components of the Multi-factor Job Priority Plugin.<br>
          # The default value for each of the following is 1.<br>
          PriorityWeightAge=1000<br>
          PriorityWeightFairshare=10000<br>
          PriorityWeightJobSize=1000<br>
          PriorityWeightPartition=1000<br>
          PriorityWeightQOS=0 # don't use the qos factor</blockquote>
        <div>We're running SLURM 18.08.8 on CentOS Linux 7.8.2003. The
          current slurm.conf is defaults as far as fair share is
          concerned:</div>
        <blockquote class="gmail_quote" style="margin:0px 0px 0px
          0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">EnforcePartLimits=ALL<br>
          GresTypes=gpu<br>
          MpiDefault=pmix<br>
          ProctrackType=proctrack/cgroup<br>
          PrologFlags=x11,contain<br>
          PropagateResourceLimitsExcept=MEMLOCK,STACK<br>
          RebootProgram=/sbin/reboot<br>
          ReturnToService=1<br>
          SlurmctldPidFile=/var/run/slurmctld.pid<br>
          SlurmctldPort=6817<br>
          SlurmdPidFile=/var/run/slurmd.pid<br>
          SlurmdPort=6818<br>
          SlurmdSpoolDir=/var/spool/slurmd<br>
          SlurmUser=slurm<br>
          SlurmdSyslogDebug=verbose<br>
          StateSaveLocation=/var/spool/slurm/ctld<br>
          SwitchType=switch/none<br>
          TaskPlugin=task/cgroup,task/affinity<br>
          TaskPluginParam=Sched<br>
          HealthCheckInterval=300<br>
          HealthCheckProgram=/usr/sbin/nhc<br>
          InactiveLimit=0<br>
          KillWait=30<br>
          MinJobAge=300<br>
          SlurmctldTimeout=120<br>
          SlurmdTimeout=300<br>
          Waittime=0<br>
          DefMemPerCPU=1024<br>
          FastSchedule=1<br>
          SchedulerType=sched/backfill<br>
          SelectType=select/cons_res<br>
          SelectTypeParameters=CR_Core_Memory<br>
          AccountingStorageHost=sched-db.lan<br>
          AccountingStorageLoc=slurm_acct_db<br>
          AccountingStoragePass=/var/run/munge/munge.socket.2<br>
          AccountingStoragePort=6819<br>
          AccountingStorageType=accounting_storage/slurmdbd<br>
          AccountingStorageUser=slurm<br>
          AccountingStoreJobComment=YES<br>
          AccountingStorageTRES=gres/gpu<br>
          JobAcctGatherFrequency=30<br>
          JobAcctGatherType=jobacct_gather/linux<br>
          SlurmctldDebug=info<br>
          SlurmdDebug=info<br>
          SlurmSchedLogFile=/var/log/slurm/slurmsched.log<br>
          SlurmSchedLogLevel=1</blockquote>
        <div>Node and partition configs are omitted above.</div>
        <div><br>
        </div>
        <div>Any and all advice will be greatly appreciated.</div>
        <div><br>
        </div>
        <div>Best wishes,<br>
          <br>
          ~Kevin<br>
          <br>
          Kevin Walsh</div>
        <div>Senior Systems Administration Specialist<br>
          New Jersey Institute of Technology</div>
        <div>Academic & Research Computing Systems<br>
        </div>
        <div><br>
        </div>
        <div><br>
        </div>
      </div>
    </blockquote>
  </body>
</html>