This version (2018/04/11 17:48) is a draft.
Approvals: 0/1
PriorityParameters      = (null)
PriorityDecayHalfLife   = 14-00:00:00
PriorityCalcPeriod      = 00:05:00
PriorityFavorSmall      = No
PriorityFlags           = FAIR_TREE
PriorityMaxAge          = 14-00:00:00
PriorityUsageResetPeriod = NONE
PriorityType            = priority/multifactor
PriorityWeightAge       = 1000
PriorityWeightFairShare = 1000
PriorityWeightJobSize   = 1000
PriorityWeightPartition = 1000
PriorityWeightQOS       = 1000
PriorityWeightTRES      = (null)
scontrol show config |grep -i prio

  • takes into account all siblings on the same level.
  • better balancing within one account, i.e. users of one account still get a high priority even if their usage is larger than their colleague users within the same account

  • shows the shares and usages of accounts and users
  • only if multifactor/priority plugin is enabled
  • reset the usage for an account:
sacctmgr modify account vsctest set RawUsage=0
  • RawShares: This is the set value when creating the account
  • RawUsage: Usage of the account user; suspected to the PriorityDecayHalfLife
  • Effctv Usage: normalized usage + siblings taken into account
  • FairShare: Factor actually used for calculating the new priority

  • priority from plugin:
PRIORITY_MF =  AGE + FAIRSHARE + JOBSIZE + PARTITION + QOS + TRES
  • only contributions if weight in slurm.conf > 0
  • total priority:
PRIORITY = PRIORITY_MF - NICE

  • lower negative values means higher priority
  • range is +/- 2147483645.
  • only privileged users can specify a negative adjustment.

Only for pending jobs:

sprio -l

Set priority explicitly (overrides nice and multifactor):

scontrol update job <job_id> priority=1

Restore multifactor:

  • hold job
  • release job
  • pandoc/introduction-to-mul-cluster/02_slurm/02_scheduler.txt
  • Last modified: 2018/04/11 17:48
  • by pandoc