Overview

SeaWulf uses a fairshare scheduling system to allocate resources fairly across all users. Job priority is influenced by historical usage, job size, and requested time. Understanding how fairshare and priority work can help you reduce wait times and maximize throughput. For an introduction to the scheduler, see the SLURM Overview.

Key Concepts

• Fairshare: Users who have used fewer resources recently are given higher priority for job scheduling.
• Job Priority: Determined by queue type, number of nodes or cores requested, walltime, and fairshare score.
• Preemption: Some queues, such as the A100 GPU queue, give higher-priority access to specific groups. Jobs from these groups may start before other jobs when resources are limited, according to SeaWulf’s scheduling policies.

Tips for Maximizing Throughput

• Submit smaller jobs when possible; multiple smaller jobs often start faster than one very large job.
• Use appropriate queues that match your job’s walltime and node requirements. See Node Utilization for guidance on resource matching.
• Stagger job submissions to avoid sending many jobs to the queue simultaneously.
• Monitor your fairshare score using squeue -u <username> or other SLURM commands. For detailed job monitoring, see Job Management.
• Use job arrays for repeated tasks to reduce queue overhead and improve scheduling efficiency.