On Sat, Feb 2, 2013 at 2:03 PM, Reuti <[email protected]> wrote:
> Am 01.02.2013 um 00:10 schrieb Allan Tran: > > > Hi, I'm trying to understand RQS and set up several rules, but they > don't quite work. > > I have several queues (4 queues A, B, C and all.q). I follow this > suggestion: > > > http://www.gridengine.info/2009/03/30/evading-quota-limits-when-resources-are-available/ > > > > So I have: > > qconf -sq A has "subordinate_list all.q B C" > > qconf -sq B has "subordinate_list all.q A C" > > qconf -sq C has "subordinate_list all.q A B" > > So the subordination should only happen if the superordinated queue is > filled completely, even by a mixed user set? > For now, that is correct. I am still trying to figure out how to implement the queues for best use. I have tried something like "subordiate_list all.q=1 A B" but then serial jobs get suspended when there's still unused slots in some cases. Would you please give some suggestions? Thank you much. > > -- Reuti > > > > Each queue A, B and C also associates to system primary group A, B, C. > > My goal for example, is to user in A group can only run in A queue, but > can only have 96 slots (8 12-core nodes) max, anything over that can run > over to all.q. > > Same thing for B and C accordingly > > My rules to limit the number of slot per queue are: > > { > > name Alimit > > description limit number of slot to all users in C queue > > enabled TRUE > > limit users {*} queues A to slots=96 > > } > > > > { > > name Blimit > > description limit number of slot to all users in C queue > > enabled TRUE > > limit users {*} queues B to slots=24 > > } > > > > { > > name Climit > > description limit number of slot to all users in C queue > > enabled TRUE > > limit users {*} queues C to slots=12 > > } > > > > So far this doesn't quite work if a user submit a lot of jobs. Meaning > if A submit 8 x 12 slots job, they all running in A queue, each job runs 12 > slots in on host. All is good. > > However, if A submit more than 8, say 9 or 10 of 12 slot jobs, sometimes > the new jobs will run 1 slot in A queue and 11 slots in all.q and it spans > onto 2 nodes. > > A.q ends up with 96+ slots. This behavior stays the same on all 3 > primary queues. > > Can someone give me some advice please? I don't quite understand the > behavior. If a limit is set for queue A, it should not have more than 96 > slots taken correct? > > > > _______________________________________________ > > users mailing list > > [email protected] > > https://gridengine.org/mailman/listinfo/users > >
_______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
