TECHNOLOGY PROFILE · TECH-0247
Slurm
Workload manager for HPC clusters.
Reference link only — not reviewed this edition. This profile preserves the prepared catalog’s original descriptions. Advantages depend on the workload; comparisons are editorial, not benchmark results.
Where it has an edge
Batch queues, resource allocation and job accounting suit shared research compute better than a web-service scheduler.
Best-fit problems
Multi-user GPU/HPC research jobs.
Tradeoffs & cautions
Queue policy, modules and cluster configuration are site-specific.
What to test before choosing
Run a representative job; inspect startup, failure recovery, quotas, storage persistence, egress and total cost.
Runtime & hardware
Cloud, HPC or private clusters
Access & licensing
Commercial compute or open-source orchestration; check terms
Execution
Local / self-hosted
Lifecycle
Not independently audited
Comparison basis
Editorial problem-fit judgement; no head-to-head benchmark performed
Connected concepts
Alternatives to investigate
Kubernetes · Ray · SkyPilot