Slurm Diamond WorkflowΒΆ

In this tutorial you will learn how to run a workflow on an HPC that uses the Slurm scheduler with these torc features:

  • Job dependencies based on input and output files

  • Jobs with different compute node requirements

  • Delayed compute node scheduling

The workflow is a derivation of the Diamond Workflow. Follow the same steps except for these deviations:

  1. Use the workflow specification file examples/slurm_diamond_workflow.json5.

  2. Change the Slurm account name.

  3. Configure and start the workflow from an HPC login node.

  4. Instead of torc jobs run, schedule a compute node to run the jobs with

$ torc hpc slurm schedule-nodes -n 1

Torc will schedule the second node with GPUs when the postprocess script is ready to run.