HON95-wiki/hpc/slurm.md

---

title: Slurm breadcrumbs:

• title: High-Performance Computing (HPC) --- {% include header.md %}

Usage

Cluster Information and Accounting

• Show partitions: scontrol show partition [-a]
• Show partition/node usage: sinfo [-a]
• Show node capabilities: sinfo -o "%20N %10A %8c %10m %25f %25G" (example)
• Show GUI (requires X11 session/forwarding): sview
• Show accounts for user: sacctmgr show assoc where user=<username> format=account
• Show default account for user: sacctmgr show user <username> format=defaultaccount

Job Submission and Queueing

• Create a job (overview): Make a Slurm script, make it executable and submit it.
• Using GPUs: See example Slurm-file, using --gres=gpu[:<type>]:<n>.
• Submit batch job: sbatch <slurm-file>
• Start interactive job: srun <job options> [--pty] <bash|app>
• Create allocation (without connecting to it): salloc <job options>
  • Use e.g. srun --jobid=<id> --pty bash to connect to it.
• Cancel specific job: scancel <jobid>
• Cancel set of jobs: scancel [-t <state>] [-u <user>]
• Show job queue: squeue [-u <user>] [-t <state>] [-p <partition>]
• Show job details: scontrol show jobid -dd <jobid>

Example Slurm Job File

#!/bin/sh

#SBATCH --partition=<partition>
#SBATCH --time=03:00:00
#SBATCH --nodes=2
# #SBATCH --nodelist=compute-2-0-[17-18],compute-5-0-[20-21]
#SBATCH --ntasks-per-node=2
# #SBATCH --exclusive
# #SBATCH --mem=64G
#SBATCH --gres=gpu:V100:2
#SBATCH --job-name="xxx"
#SBATCH --output=log.txt
## SBATCH --mail-user=user@example.net
# #SBATCH --mail-type=ALL

# Run some program on one processor on all nodes
srun uname -a

{% include footer.md %}