Someone is hogging the cluster with lots of long jobs. What can I do?

This is where Partitions come in handy. The short version: If your jobs run only for 4 hours or less, you can use the high_prio QoS, either by adding --qos=high_prio to your srun command or the submit script. Or if you use plus_slurm or plus_slurm_matlab, you are going to find a QoS parameter there.

If you use the high_prio QoS:

1. Your jobs get priority over all jobs in the normal (default) queue.
2. Your jobs can even preempt (i.e. stop) jobs in the normal queue.
3. But: You are limited to 50% of the cores.

How can I change my default shell?

As we use FreeIPA for our User Management, you cannot use chsh to change your default shell. Instead, you can use the ipa command:

ipa user-mod --shell=/bin/bash USERNAME

Starting a (interactive) job in a container is really slow. What can I do?

This is especially important for Matlab users as they need to use containers in the current setup.

The problem is that container images are rebuilt quite often and pulling these large images from the web takes time. To mitigate this, you can pull the image you need into your homefolder once. For instance, for the Matlab container, you can do this:

apptainer pull ~/matlab.sif oras://ghcr.io/thht/obob-singularity-container/xfce_desktop_matlab:latest

This is going to download the container into your home folder as matlab.sif.

Now you can use the path to the local image when you start your interactive session or job:

• When using the OOD web interface, just enter /home/USERNAME/matlab.sif as the "Apptainer Image".
• When using plus_slurm_matlab, you can specify the image path in the cfg.container option.