Slurm Job Scheduling System

The principle of a resource manager is to allocate compute cores according to your needs.

You interact with the resource manager using the following commands:

• sbatch: submit a job to a queue (called partitions in Slurm)
• scancel: cancel a job
• squeue: view running or pending jobs
• sinfo: view queue/partition status
• srun: run a command immediately

How to Launch a Job

To launch a job, create a script where you request resources and call your program. Example:

#!/bin/bash

#SBATCH -J test # Job name

#SBATCH -o job.%j.out # Stdout output file (%j = job ID)

#SBATCH -N 2 # Number of nodes requested

#SBATCH -n 16 # Number of MPI tasks

#SBATCH -t 01:30:00 # Run time (hh:mm:ss)

# Launch the MPI-based executable

prun ./a.out

Submit the script with the sbatch command:

[test@sms ~]$ sbatch job.mpi

Submitted batch job 339

The system will return a job ID, which you can use with commands like scancel or squeue.

Monitoring Job Status

You can monitor the status of your job using:

• List all active jobs:
• squeue
• Show detailed status of a specific job:
• scontrol show job <jobID>