Argon HPC

Getting Started

Creating a Job Script

Below is an example of a typical job script for submitting jobs to the Argon HPC cluster. Each line is explained in detail.

#!/bin/sh
#$ -cwd
#$ -pe 128cpn 512
#$ -q UI-MPI
#$ -l h_vmem=80G
#$ -l mem_free=256G
#$ -M email@uiowa.edu
#$ -m beas
#$ -o file.log
#$ -e file.log
#$ -V
      

Explanation of the Script

• #!/bin/sh: Specifies the shell interpreter for the script.
• #$ -cwd: Ensures the job runs in the current working directory.
• #$ -pe 128cpn 512: Requests a parallel environment with 512 cores, each node having 128 cores per node.
• #$ -q UI-MPI: Specifies the queue to submit the job to (in this case, the UI-MPI queue – note: 48 hr timeout). UI queueing avoids timeout problem.
• #$ -l h_vmem=80G: Specifies the hard memory limit per core or slot.
• #$ -l mem_free=256G: Ensures the job runs only on nodes with at least 256GB of free memory available.
• #$ -M email@uiowa.edu: Sets the email address for job notifications.
• #$ -m beas: Specifies when to send email notifications:
  • b: When the job begins.
  • e: When the job ends.
  • a: If the job aborts.
  • s: If the job is suspended.
• #$ -o file.log: Specifies the file to write standard output.
• #$ -e file.log: Specifies the file to write standard error.
• #$ -V: Exports all environment variables to the job environment.

Useful Commands

Submitting a Job

qsub [jobfile_path]

Use qsub followed by the path to the job file to submit a job to the cluster.

Checking Job Status

qstat

Use qstat to check the status of submitted jobs. Add your username to filter by your jobs:

qstat -u 

Deleting a Job

qdel [job_id]

Use qdel followed by the job ID to delete a job from the queue.

Releasing a Held Job

qrls [job_id]

Use qrls to release a job that is on hold.

Loading Modules

module load [module_name]

Use module load to load specific software or libraries. For example:

module load python

Listing Loaded Modules

module list

Displays all currently loaded modules.

Unloading a Module

module unload [module_name]

Unloads a specified module.

Troubleshooting Tips