What is SLURM?
SLURM (Simple Linux Utility for Resource Management) is an open-source workload manager and job scheduler used on SeaWulf to manage compute resources and run jobs efficiently across the cluster.
How SLURM Works
SLURM serves three primary functions on SeaWulf:
Resource Allocation: Allocates exclusive or shared access to compute nodes for specified durations.
Job Execution: Provides framework for starting, executing, and monitoring jobs on allocated nodes.
Queue Management: Manages job queues and arbitrates resource contention between competing jobs.
Getting Started
Before using SLURM commands on SeaWulf, load the SLURM module:
module load slurm
Essential SLURM Commands
Here are the core commands you'll use to submit and manage jobs:
| Function | Command | Description |
|---|---|---|
| Submit batch job | sbatch [script] |
Submit a job script to the queue |
| Interactive job | srun --pty bash |
Start an interactive session on compute node |
| Check job status | squeue |
View current job queue and status |
| Check your jobs | squeue --user=$USER |
View only your jobs |
| Cancel job | scancel [job_id] |
Cancel a running or queued job |
| Job details | scontrol show job [job_id] |
Show detailed job information |
| Node information | sinfo |
Display node and partition information |
Quick Examples
Check your running jobs:
squeue --user=$USER
View details about a specific job:
scontrol show job 123456
See available partitions and nodes:
sinfo
Job Script Basics
SLURM jobs are defined using job scripts that specify resource requirements and commands to run. Job scripts contain special directives that begin with #SBATCH.
Essential SLURM Directives
| Resource | Directive | Example |
|---|---|---|
| Job name | #SBATCH --job-name= |
--job-name=my_job |
| Number of nodes | #SBATCH --nodes= |
--nodes=2 |
| Tasks per node | #SBATCH --ntasks-per-node= |
--ntasks-per-node=40 |
| Memory per node | #SBATCH --mem= |
--mem=64GB |
| Wall time | #SBATCH --time= |
--time=02:30:00 |
| Partition/Queue | #SBATCH -p |
-p short-40core |
| Output file | #SBATCH --output= |
--output=job_%j.out |
| Error file | #SBATCH --error= |
--error=job_%j.err |
%j placeholder in filenames is automatically replaced with the job ID, making it easy to track output from multiple jobs.Useful Environment Variables
SLURM automatically sets several environment variables that your jobs can use:
| Variable | Description |
|---|---|
$SLURM_JOBID |
Unique job identifier |
$SLURM_SUBMIT_DIR |
Directory where job was submitted from |
$SLURM_JOB_NODELIST |
List of nodes allocated to the job |
$SLURM_NTASKS |
Total number of tasks for the job |
$SLURM_CPUS_PER_TASK |
Number of CPUs allocated per task |
$SLURM_JOB_NAME |
Name of the job |
Example usage in a script:
echo "Job ID: $SLURM_JOBID" echo "Running on nodes: $SLURM_JOB_NODELIST" cd $SLURM_SUBMIT_DIR
Best Practices
Resource Estimation
Request only the resources you actually need. Over-requesting resources can lead to longer queue times and reduced system efficiency.
Output Files
Always specify output and error files to capture job information. Use %j to include the job ID in filenames:
#SBATCH --output=job_%j.out #SBATCH --error=job_%j.err
Module Loading
Load all required modules within your job script to ensure consistent environments across compute nodes. Always start with module purge to avoid conflicts.
Test First
Test your scripts with small resource requests first. Use interactive sessions to debug before submitting large batch jobs.
Need Help?
For detailed SLURM documentation, visit the official SLURM documentation. For SeaWulf-specific questions, submit a ticket to the IACS support system.