Am 06.04.2011 um 10:00 schrieb Vic: > Hi Stephen, > >> I think thats what I was getting at but semantically different. > > Yes - it's that semantic difference that's the trouble :-) > >> The low queue is the borrow queue which can get onto any box but will >> give way immediately if there is a hi job which belongs to either >> department. > > That's not going to work in our environment; the jobs aren't > differentiated by priority, as they're all the same. So either everything > gets submitted to the low queue - meaning there is no preemption at all - > or everything gets submitted to the high queue - meaning that there is no > sharing between the groups. > > Neither of these scenarios improves on just having two grids with no > sharing :-(
Can you specify what you mean in detail by the owner group should have access to it's nodes? Should already running job from the other group being killed? I could think of a cron job which changes either the user_lists when there are waiting job of the owner of this part of a cluster, or adjust a slot limit in an RQS in a similar way: limit users @group1 hosts @group2 to slots=0 to block further starts. Already running job would continue. -- Reuti > Vic. > > _______________________________________________ > users mailing list > users@gridengine.org > https://gridengine.org/mailman/listinfo/users _______________________________________________ users mailing list users@gridengine.org https://gridengine.org/mailman/listinfo/users
