FairShare and Priority
Slurm uses a fairshare system to balance cluster usage across labs and users over time.
What FairShare Does
In simple terms, labs that have used less of their expected share of the cluster tend to get higher priority, while labs that have recently used more tend to get lower priority.
What Affects Priority
- Historical cluster usage
- Account or lab shares
- Requested resources
- Partition and QoS rules
Slurm supports fairshare and hierarchical priority controls through associations, partitions, limits, and Quality of Service (QoS).
Check FairShare
sshareTypical Interpretation
| FairShare Score | Interpretation |
|---|---|
| Near 1 | High priority; account has used relatively little computing power recently. |
| Around 0.5 | Moderate priority |
| Near 0 | Low priority; account has recently used a large share of the cluster resources. |
Practical Advice
- Request only the CPUs, memory, and GPUs your job actually needs to compute efficiently.
- Use automated batch jobs instead of leaving idle interactive sessions running.
- Utilize job arrays for repeated work to reduce cluster scheduling overhead.
- Do not request long walltimes unless explicitly needed, as short walltimes track cleanly into free windows.