How-to Submit an Array Job

Step-by-step guide

While SLURM is not designed for HTC (High Throughput Computing) workloads, the use of job arrays can facilitate HTC like job submissions. To launch an array of jobs:

1. Set the "array" option in your SLURM script
2. Use the resulting integer values to loop over your simulations, either by reading the integer into your executable as an input or by translating its value into to filenames, directories, etc.

Let's look at a simple example:

#!/bin/bash
#SBATCH --partition=compute    ### Partition
#SBATCH --job-name=ArrayJob    ### Job Name
#SBATCH --time=00:10:00        ### WallTime
#SBATCH --nodes=1              ### Number of Nodes
#SBATCH --ntasks=1             ### Number of tasks per array job
#SBATCH --array=0-19           ### Array index
#SBATCH --account=hpcrcf       ### Account used for job submission

echo "I am Slurm job ${SLURM_JOB_ID}, array job ${SLURM_ARRAY_JOB_ID}, and array task ${SLURM_ARRAY_TASK_ID}."

 Let's look at the output of this job:

[duckID@ln1 array]$ sbatch array.srun 
Submitted batch job 20303
[duckID@ln1 array]$ cat slurm-20303_1.out 
I am Slurm job 20305, array job 20303, and array task 1.
[duckID@ln1 array]$ cat slurm-20303_19.out 
I am Slurm job 20323, array job 20303, and array task 19.
[duckID@ln1 array]$ 

Note that all jobs in the array share the same SLURM_ARRAY_JOB_ID but have a different SLURM_JOB_ID, i.e. they're separate jobs as far as SLURM is concerned. Most importantly, note the value of SLURM_ARRAY_TASK_ID corresponds directly to the array values we specified in our SLURM script. It's this value we will use to communicate to the job which specific job within the array of jobs that it's running.

SLURM will run the constituent jobs as resources become available.  It's even possible that they might all start simultaneously.  See the SLURM docs if you need to limit parallelism.

Related articles