RCC users can run jobs either interactively or by submitting jobs to be scheduled to run on allocated resources (CPU time, memory, etc.) by a resource manager. The RCC uses SLURM (Simple Linux Utility for Resource Management) to schedule jobs to run on its compute clusters (Midway2, Midway3, Beagle3, GM4, MidwayR, etc.). In addition to learning the best practices, this workshop will provide users with a clear understanding of all the compute partitions available at the RCC to which a job can be submitted, how to create a Slurm batch script to submit a job, how to configure a Slurm job, how to use Slurm commands, how to submit a job and how to avoid common mistakes that usually result in a job waiting for a long time in the queue before running or failing to run.

Objectives:

Participants will learn:

various RCC compute resources and partitions for running jobs
slurm commands, how to create a Slurm batch script and how to submit batch jobs
RCC module system and run time environments
how to submit serial single processor and parallel (OpenMP and MPI) multiple processor jobs
how to submit GPU jobs
how to submit a job that is carried on several times by a given code, differing only in the initial value of some high-level parameter for each run ( Slurm job array)
how to pack jobs and schedule independent processes inside a Slurm job allocation
how to submit Message passing parallel jobs (MPI), multi-threading (OpenMP) and hybrid jobs.
how to submit a job arrays which provide a convenient way to submit a large number of independent jobs
how to request a Slurm interactive session
best practices and how to debug your Slurm script

Level: Introductory

Duration: 1.5 hours

Github repository: https://github.com/rcc-uchicago/Intro_to_SLURM

Job scheduling on SLURM - Best Practices