HPC Job scheduler Pbs Torque Tricks Tips


Torque - Tera Scale Open Source Resource & Queue Manager
Torque Architecture consist of
a)Torque server : run on master node. control submission and running of jobs.
b)PBS scheduler: take care of scheduling job across the cluster.
c)mom daemon: star and stopping job on client nodes.

Job submit Script File #PBS and torque script file.
-N Name
-M Mail id
-l Job Resource
-O name
-e error file name
-q queue
#PBS -N "Job Name"
#PBS -l "resource"
#PBS -l ncpus=25
#PBS -l mem=256mb
#PBS -l Walltime=3:20:00
#PBS -o outputfilename.out
#PBS -e errorfile.err
#PBS -q job-queue
#PBS -s "specify shell type"

For More Information.
Torque submit and Monitoring Job

Other Useful Information.
1)qmgr -c "s s job_history_enable =t"
2)qmgr -c "s s pbs_license_info = 6200@license-server-ip-address"
3)qmgr -c "create node compute-node-hostname"

Queue Limit
1)max_queued = [u:PBS_GENERIC=2] 
Maximum user can run two jobs.
2)max_run_res.ncpus = [g:engr=80]

engr group member maximum can use up to 80 cpu.

No comments:
Write comments