Slurm Job Submission Guide (Performetrica HPC)

Documentation Status

Overview

This guide explains how to submit, monitor, and troubleshoot jobs on the Performetrica HPC cluster using Slurm.

Preparing a Job Script

Example scripts may be located under:

/perf/shared/

Copy and modify:

mkdir -p $HOME/workfolder
cd $HOME/workfolder
cp /perf/shared/example_submit.sh my_experiment.sh
vi my_experiment.sh

Submitting a Job

sbatch my_experiment.sh

Minimal Example

#!/bin/bash
#SBATCH --job-name=test_job
#SBATCH --partition=defq
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=4
#SBATCH --mem=8G
#SBATCH --time=01:00:00
#SBATCH --output=%x-%j.out

echo "Running on $(hostname)"

Basic Commands

sbatch job.sh
squeue
squeue -u $USER
scancel <jobid>
sinfo
scontrol show job <jobid>
sacct -j <jobid>

Important Notes for Performetrica HPC

  • Default partition: defq
  • QoS is NOT used in this cluster
  • Always request only necessary resources
  • Prefer %x-%j.out for outputs

OpenMP Example

#!/bin/bash
#SBATCH --partition=defq
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=16
#SBATCH --mem=16G
#SBATCH --time=02:00:00

export OMP_NUM_THREADS=$SLURM_CPUS_PER_TASK
./my_openmp_program

MPI Example

#!/bin/bash
#SBATCH --partition=defq
#SBATCH --nodes=2
#SBATCH --ntasks-per-node=8
#SBATCH --time=01:00:00

srun ./my_mpi_program

Monitoring

squeue -u $USER
sacct -j <jobid>

Best Practices

  • Match ntasks vs cpus-per-task correctly
  • Set OMP_NUM_THREADS
  • Avoid oversubscription
  • Optimize total time (queue + runtime)

Available Software

module avail