Scheduling & Dispatch Policy
This guide will help users understand how their jobs get executed and why they get the priority that they do.
First-In, First-Out Scheduling
The SLURM installation at USF implements a First-In, First-Out (FIFO) scheduling system, in relation to resources requested and Quality of Service (QOS) modifier. So, in practice:
- if two identical jobs (requesting identical resources/run-times) are submitted to SLURM, the first job submitted will begin running before the second job submitted.
- if two differing jobs are submitted at the same time (and cluster resources are close to maximum utilization), the “smaller” job (smaller in terms of requested resources and/or run-time) will be dispatched before the “larger” job (larger in terms of requested resources and/or run-time).
- if two identical jobs (requesting identical resources/run-times) are submitted to SLURM, but one job has a higher QOS modifier, the job with the higher QOS will be dispatched before the job with the lower QOS.
- For information on Preemption, please see SLURM Preemption Guidelines