SLURM

SLURM

Slurm is a batching system that distributes jobs to several machines respecting given constraints. At the moment we use it to run simulations and stuff on our compute servers i1 - i5.

Usage

Even though Slurm is able to schedule jobs of different users, ask your colleagues whether it is ok to put more jobs to the machines The CECI has a good tutorial. Another good resource for job arrays can be found here.

Partitions and Features

Partitions and features are used to decide on which machines jobs are executed. We have the following partitions:

• i: all machines (i1-5)

• fastsingle: good single-core performance (i3)
• fastmulti: good multi-core performance (i4, i5)
• huge: lots of RAM (i4, i5)

Important Commands

• sinfo: shows which machines are free to use and other status things
• squeue: shows all tasks in the queue and in execution