Batch system: Submitting jobs with SLURM

The queueing system is based on SLURM, an open source, fault-tolerant and highly scalable cluster management and job scheduling system for Linux clusters.

Slurm has three key functions:

1. It allocates exclusive and/or non-exclusive access to resources (compute nodes) to users for some duration of time.

2. It provides a framework for starting, executing and monitoring jobs on the set of allocated nodes.

3. It arbitrates contention for resources by managing a queue of pending work.

For more information about SLURM you can consult the guides and tutorials available on their website: Quickstart user guide and SLURM tutorials. There’s also some useful links at portalusuarios.cesga.es

Index of contents:

• Memory and time
• Partitions
• QOS
  • User Account Limits
  • Assoc-Group Limits
• Basic commands
  • sbatch
  • srun
  • sinfo
  • squeue
  • scancel
  • scontrol
  • sacct
  • seff
  • sqstat
• Sample Job Scripts
  • Using srun
  • Using sbatch
  • Using sbatch and GPUs
  • OpenMP job submission
  • MPI job submission
  • Hybrid MPI/OpenMP programs
• Job array
• Multiple tasks on a job
  • Running multiple programs with --multi-prog option
  • Running multistep programs and -r, –relative=<n>
  • Running multiple programs within a job using GNU Parallel and srun
• Tasks on specific cores
  • 1. Task level affinity (Pure MPI)
  • 2. Thread level affinity (OpenMP and MPI/OpenMP jobs)
• Job states
• Output files
• Email notifications