1. What is a scheduler?

Why HPC clusters need a scheduler

An HPC cluster typically has hundreds or thousands of compute nodes shared by hundreds of users. A scheduler decides which job runs on which node and when, so resources are allocated fairly and predictably.

On NRIS clusters (Saga, Olivia, Betzy, Fox) the scheduler is Slurm. Other clusters may use PBS, LSF, or others — the syntax differs, the concepts do not.

Three things to take away

• Workloads on the cluster are submitted as jobs; a job is just a shell script.
• The login node is for editing scripts and submitting them — not for running workloads.
• Without a scheduler, one runaway user could starve everyone else on the cluster.

Read the canonical section →