The job scheduler SLURM version 22.05.8 is installed on the OMNI cluster. SLURM is used to distribute the users' computing jobs on the cluster in such a way that it is used as efficiently as possible and the waiting times are as short as possible. More information on setting up jobs can be found here. The SLURM documentation can be found here.

Tip: SLURM is only available if the SLURM module is also loaded(module load slurm). By default, the SLURM module is always loaded at login, but it can be unloaded accidentally or intentionally by user actions (e.g. with module purge). If you do not have the SLURM commands available, use module list to ensure that the module is loaded. Further information on modules can be found here.

A SLURM job obtains its settings from various sources. These are in descending order of priority:

• Command line parameters of the command used to set the job.
• Environment variables, if set.
• The parameters at the start of a job script, if specified.
• The standards set in SLURM.

This means, for example, that you can specify in a job script which settings it should run with, and if you need different settings once, you can simply specify a command line parameter when calling it.

Terms

In principle, SLURM recognizes and takes into account the division of the cluster into nodes with individual cores. When setting up jobs, there are several options for requesting resources. In SLURM terminology:

• A job, a self-contained calculation that can consist of several tasks and is assigned(allocated) certain resources such as individual CPU cores, a certain amount of RAM or complete nodes.
• A task, the run of a single process. By default, one task is started per node and one CPU is allocated per task.
• A partition (often also called a queue) is a queue in which jobs are placed.
• A CPU in Slurm refers to a single core. This differs from the usual terminology in which a CPU (a microprocessor chip) consists of several cores. Slurm speaks of "sockets" when referring to CPU chips.

Console commands and options

Here are the most common commands that you will need as a SLURM user.

Many of these commands have options (command line parameters) that you can specify when calling them. The most important ones are briefly listed here:

You can find a full list for each command using man or --help and also in the SLURM documentation of sbatch.

Environment variables

SLURM uses a number of environment variables. You can use some of them to give SLURM jobs settings (unless you overwrite them when setting up the job using command line parameters), they are described in the sbatch documentation in the section on input variables section. Other variables are set by SLURM when the job is set up. You can use these within the job to query job settings. Only a few examples are listed here, the section on output variables in the sbatch documentation contains a complete list.