Our HPC systems make use of Slurm's Multifactor Priority system to evaluate what job to run next. Jobs are selected based on your group's "fairshare" priority, the partition you submitted to, the size of the job (number of nodes), and the length of time the job has been queued for.The "fairshare" priority system has the most influence on determining when your job will start. The fairshare weight assigned to your job is a function of your historical use of the cluster resources, weighed against the proportion of the priority assigned to your group. In simple terms, your jobs are assigned a higher priority if your group's usage falls below the proportion assigned, and is lowered if you've exceeded that proportion.
Current Settings
Below are the current SLURM settings in effect on Pegasus. These are subject to change at the discretion of the HPC Systems staff and the HPC Advisory Committee.
PriorityType=priority/multifactor PriorityDecayHalfLife=14-0 PriorityUsageResetPeriod=NONE PriorityFavorSmall=yes PriorityWeightFairshare=100000 PriorityWeightAge=10000 PriorityWeightPartition=100000 PriorityWeightJobSize=10000 PriorityMaxAge=1-0
