TTU Home IT Division HPCC

Batch Queuing

LSF Batch Queue



We do not allow any user jobs (Testing/Production) on frontend under any circumstances

Hrothgar uses the LSF queuing system. All jobs which run more than 2 minutes and all multiprocessor jobs should be submitted to LSF.

Instead of an interactive command like

./myjob

use

bsub -q serial -n 4 < my.sh

The components of this command are:

  • bsub : submit to the lsf system
  • -q serial: a bsub switch to select the lsf queue serial .
  • bqueues to list the available queues.
  • -A: followed by the name of the executable to run.

    Usually you will want to use a command script with bsub. The command script for the previous example would be


    #!/bin/bash [or tcsh]
    #BSUB -q serial [or idle or parallel]
    #BSUB -n 1
    #BSUB -r "span[ptile=2]" [2 cpu/node. Please add]
    #BSUB -J myjobname [optional for bjobs]
    #BSUB -o myjob.out [or %J.out is job#.out]
    #BSUB -e myjob.err [or %J.err]
     ./a.out > test.out

    I f you create this script as say runjob using a text editor, then bsub < runjob would be equivalent to bsub -q serial -np 4 < myjob . Don't bsub runjob , which will, submit the job to idle quque. The LSF queues at HPCC, in order of decreasing priority, are:

  • 48Hserial: serial jobs, with total runtime less than 48 hours.
  • 48Hquadpar: parallel jobs that run less than 48 hours and run on quad core node.
  • 48Hpar: parallel jobs that run less than 48 hours.
  • 2WKpar parallel jobs that run for more than 48 hours and less than 2 weeks.
  • 2WKserial serial jobs, that run for more than 48 hours and less than 2 weeks.

    Each queue, in order of decreasing priority, will run all the jobs in that queue, up to the queue limit or the machine limit, whichever is less.
  • Jobs in idle queues will be suspended if the machine limit is reached before the queue limit.

    Click here to access the ppt document for additional information about parallel job submission, compile optimizations