Running Jobs on OnDemand
ATTENTION: OnDemand is in pre-production. View or subscribe to SDSC News for updates. To request access to this resource, contact consult@sdsc.edu.
Users can submit interactive or batch compute-intensive jobs from the front-end node, rocks-131.sdsc.edu. Jobs are submitted to the Sun Grid Engine (SGE).
Running Batch Jobs
Batch-parallel job submission to the SGE uses the following command:
% qsub myjob_script
The script, myjob_script, looks like the following:
#!/bin/sh
#
#$ -N myjob_name
#$ -S /bin/sh
#$ -o
myjob.output
#$ -e myjob.error
#$ -M myemail@address.edu
#$ -m es
#$ -pe mpi 8
#$ -q all.q
#
cd /my/current/directory
mpirun -np $NSLOTS -machinefile $TMPDIR/machines ./a.out
This script requests 8 cores which can be obtained from 2 ondemand nodes. This example also uses the default all.q versus on_demand.q for preemptible jobs which need immediate start. The qstat command will display information about the job queues and the running jobs. The qdel command can cancel your queue.
Running Interactive Jobs
Users can obtain interactive nodes from the SGE at rocks-131 .sdsc.edu. For example if you are requesting 8 processors (cores), do the following:
% qrsh -pe mpi 4 -l h_rt=01:00:00 : requsting 8 cores for 1 hour
The SGE will assign you the nodes and log you into the one of the first compute nodes that you are assigned to.
If you are asked for your password, you might have provided a password on the ssh-keygen process at the first time of login. You can redo the ssh-keygen to set passwordless ssh.
Execute your job with the following command:
% mpirun -np 4 -machinefile $PE_HOSTFILE ./a.out
Log off the current interactive node to quit the interactive session. Some uses of this resource are debugging, visualization, post-processing data, and interactive software.
