Your contribution level, or fairshare, has some affect on how much of the cluster you can effectively use at any given time. Larger account fairshares generally result in shorter average job wait times in the idle queue. All else being equal, if your account usage is regularly above fairshare, the average job wait times will be longer compared to when your account is well under fairshare. If you are concerned about long wait times, check your current usage versus your fairshare to see whether this is a factor.
-
To see fairshare and current usage for all users and groups (see also this FAQ), type the command:
mdiag -f -
You can "grep" to see the output for just your user or group account:
mdiag -f | grep account_name -
The following is an example of the tabulated output:
FairShare Information Depth: 7 intervals Interval Length: 2:00:00:00 Decay Rate: 0.75 FS Policy: DEDICATEDPS (usage is tracked by processor seconds dedicated to each job) System FS Settings: Target Usage: 0.00 Flags: 0 FSInterval % Target 0 1 2 3 4 5 6 FSWeight ------- ------- 1.0000 0.7500 0.5625 0.4219 0.3164 0.2373 0.1780 TotalUsage 100.00 ------- 24748.7 53620.5 46074.5 49319.7 46570.2 49744.9 55666.6 account_name 13.97 8.69+ 13.11 11.27 10.56 8.45 20.62 29.97 20.81 Col. 1: Account name
Col. 2: Weighted percentage [ ∑i=0,6 (wi pi) ÷ ∑i=0,6 wi ] of the cluster used in the last two weeks by account_name
Col. 3: Target fairshare percentage of the cluster, e. g., 8.69% means account_name has bought in at the level of 130 cpu's, which is 8.69% of the current 1496 compute processors in the cluster. N. b., the "+" after target fairshare, in this case 8.69+, means this account is allowed to run at a level above the fairshare
Col. 4-10: Percent usage over two weeks in two-day increments (Row 1). The most recent two-day period is "0"The scheduler determines the weighted average usage, 13.97%, by an algorithm that weights (Row 2) the last seven two-day periods with a decay rate of 75% for each period, so the most recent usage has the largest impact. This means if a group has not used the cluster within two weeks, its usage is zero when setting job priorities.
-
To see your usage as a percentage of your group account:
mdiag -f -o account --flags=relative
ACCRE staff also monitor the scheduler with the goal of balancing fairness with usage. Please also see our buy-in policy. To better understand your particular fairshare allocation, an account PI may at any time consult ACCRE Administration.
