Running jobs using Slurm#

In order to run jobs on the Esrum cluster, you must connect to the head node (see Connecting to the cluster) and queue them using the Slurm job management system. Slurm takes care of automatically queuing and distribute jobs on compute and GPU nodes when the required resources are available.

While it is permitted to run small jobs directly on the head node, more resource intensive jobs must be queued using Slurm. We will terminate jobs running on the head node without prior warning, if it is necessary to prevent them from impacting users of the cluster.

The documentation for running jobs is split into three sections:

  1. The Basic Slurm jobs section describes the basic procedure of queuing jobs using the sbatch command, how to request resources for your jobs, how to monitor your jobs, and how to run a shell on a compute node when you need to work interactively.

  2. The Advanced Slurm jobs section describes additional ways to monitor your jobs and the cluster as a whole, how to batch multiple, similar jobs using sbatch, how to use srun for executing singular commands, and more.

  3. The Using the GPU / high-memory nodes section describes how to run jobs on the GPU / high-memory node, including best practices for using this limited resource.