Policies related to usage of this resource via Grid services
- Fairsharing Procedure (2014 (June onwards)
- 2 nodes reserved for Local and SAgrid only.
- The rest are then divided up as follows
- The fair share policy now is set to calculate over 7 days.
- +CLASSCFG[dteam] FSTARGET=10 MAXPROC=10,30
- +CLASSCFG[saatlas] FSTARGET=20 MAXPROC=10,255
- +CLASSCFG[batch] FSTARGET=50 MAXPROC=10,255
- +CLASSCFG[sagrid] FSTARGET=10 MAXPROC=10,64
- +CLASSCFG[ops] FSTARGET=10 MAXPROC=10,30
- The FSTARGET is the % of the cluster one can use if all cores are in use.
- The MaxProc are the lower and upper limit of CPU's that can be used
- Atlas and local users should be able to use almost all the cores available
- (They cant use all, other wise Bruce's SAGRID monitoring jobs will fail).
- Local users are weight 50 % compared to 20% for ATLAS users.
- If this works, Chris Lee plans to add the 2 reserved nodes back into the queue so we are not wasting resources when users are not available
- Jobs are also not stopped if somebody has higher priority, so when a user tries to suddenly use 1/4 of the cluster as was the case here now, they are going to wait a bit.
- Policy Document from the UJ Easter Recess Workshop on 10.4.2009
- OSG Site Usage Policy - draft policy extensions
- SAGrid/gLite Site Usage Policy - to be defined
