Submitting Jobs at HPC facility

Jobs on the HPC cluster must be submitted using the PBS script.

Under NO circumstances should you execute your job directly on the Head node or any compute node.

All the jobs are to be submitted by scheduler not directly.

Some useful openPBS commands to submit and manage jobs

Job scripts

Model scripts for submitting jobs can be found on the Scripts page

Serial job: Serial_script.sh
Parallel job: OpenMPI_script.sh, OpenMPIsub_script.sh, OpenMPIch2sub_script.sh, Parallel Omp_script.sh
Cuda: Cuda_script.sh
Gaussian: gaus_script.sh

Submitting jobs

qsub : Submit a job to the default queue (submit)
qsub -q : Submit a job to a particular queue
qsub -I : Submit a job for an interactive run (will return a prompt when job is run)
qsub -h : Submit a job and put a hold on it (useful for starting jobs with dependencies)

Checking job status

qsub : qstat -q : List jobs in a specific queue
qstat -u :List jobs for a particular user

Other job-related commands

qrls : release a hold on a job
qdel : stop and remove a job from pbs
pbsnodes: List the status of all the compute nodes

Instructions for Submitting Gaussian runs

Note: The Gaussian users should ask for access to the Gaussian package, users by default do not have access to it. Access can be requested at hpc@iitmandi.ac.in

Login to server with -Y option to enable X forwarding

For the new cluster

Login to server with -Y option to enable X forwarding
ssh -Y username@10.8.1.17
Step 0: Add following lines to end of your /home/username/.bashrc file
export PATH=$PATH:/opt/soft/share/g09
export PATH=$PATH:/opt/soft/share/g09/gv
export GAUSS_EXEDIR=/opt/soft/share/g09
export g09root=”/opt/soft/share/g09″
export GV_DIR=”${g09root}/gv”
to access ghostview type
$gv

Do not submit the Gaussian runs from ghostview window

After preparing the input file, i.e., the —-.com file, specify number of threads required at the beginning of this file – alternatively you can do it from the gv window %nprocshared=4

Note that Gaussian is inefficient in running in parallel, for large systems use maximum of 8 threads, usually 2-4 threads are sufficient.

Model script of submitting Gaussian programs can be downloaded from here http://10.8.1.15/gaus.sh

Instructions to be used in the submit script downloaded

run the program with same number of processors specified in the —-.com file – in case of
%nprocshared=4
use
#PBS -l ncpus=4
Please make sure that these two numbers are equal for efficient utilization of available resources and give correct information to the scheduler.
finally submit the script for running using
$qsub gaus.sh