SQS: Slurm Quick Start

Quick start to SLURM.

Custom software

1. Clone image

Clone new Enroot image from existing images. See example.

After calling below command, an interactive shell spawns, where you can install your required libraries. When you're finished close the shell via exit. The cloned image will be created and is available under /netscratch/$USER/.

Format:

bash clone_image.sh <postfix>

Example:

bash clone_image.sh v1

Note:

• This script uses as template image dlcc_pytorch_20.10.sqsh. Find more images under /netscratch/enroot/.
• <postfix> is a custom identifier which allows you to make your image unique.

2. Modify image

Modify your own Enroot image. See example.

Same procedure as for image cloning.

Format:

bash modify_image.sh <postfix> <new_postfix>

Example:

bash modify_image.sh v1 v2

Note

• <postfix> is the custom identifier specified during image cloning.
• <new_postfix> is the postfix of your modified image. (leave same as <postfix> for overwrite)

Running jobs

1. Simple single run

Start a single-GPU job. See example.

This script requires to be set in a directory scripts next to the script which is to be called.
The arguments <args> passed with start.sh will be passed to your run.sh and can be accessed accordingly via $1, $2, ....

Format:

bash start.sh <args>

Example:

bash start.sh -a 1 -b 2 -c 3

Note

• run.sh:

run.sh requires the command for calling your script.

• start.sh:

start.sh requires variables PARTITION and IMAGE.

• PARTITION: See possible partitions, e.g., RTX3090.
• IMAGE: See possible images under /netscratch/enroot/, e.g., dlcc_pytorch_20.10.sqsh.

• wrapper.sh:

wrapper.sh is implicitely called and required for running the script in the VM instead within cluster itself.

Additional resources

http://projects.dfki.uni-kl.de/km-publications/web/ML/core/hpc-doc/

https://support.ceci-hpc.be/doc/_contents/QuickStart/SubmittingJobs/SlurmTutorial.html