hi may be just switch or fabric no need both for jobs only to span single switch regards
Sent from my iPad On Jul 11, 2012, at 19:33, Brian Smith <b...@usf.edu> wrote: > One problem I'm seeing with this approach is that if I want to allow users to > select processors per node, the number of pes that will need to be defined is > quite large. > > I was thinking that I could create a complex of type STRING, say "fabric" > where we have values "a" or "b" defined as a complex value for each host. > Then, have a STRING, say "switch", with some value "a", "b", ... etc. > > This would allow us to select: > > hard request: fabric=* > soft request: switch=* > > When you use a wildcard with a PE name, only ONE PE is selected for the job. > In the case of the wildcard selected string, would only a single string > instance be selected or would any matching string be picked? > > If only a single instance per job, say > > fabric=a > switch=b > > gets picked, then I could set > > -hard -l fabric=* -soft -l switch=* > > in the sge_request file to pull off the desired behavior, and keep only a > small set of PEs available. > > -Brian > > Brian Smith > Sr. System Administrator > Research Computing, University of South Florida > 4202 E. Fowler Ave. SVC4010 > Office Phone: +1 813 974-1467 > Organization URL: http://rc.usf.edu > > > On 07/11/2012 04:28 PM, Reuti wrote: >> Am 11.07.2012 um 22:13 schrieb Brian Smith: >> >>> Hi, all, >>> >>> Is the following formatting permitted for the pe_list attribute in a queue >>> configuration: >>> >>> ... >>> pe_list [@hgrp1=pe_sw1],[@hgrp2=pe_sw2],[@hgrp3=pe_sw3],[@allHosts=pe_all] >>> ... >>> >>> or, perhaps >>> >>> ... >>> pe_list [@hgrp1=pe_sw1 pe_all],[@hgrp2=pe_sw2 pe_all],[@hgrp3=pe_sw3 >>> pe_all] >>> ... >> >> Besides the missing default value (which could even be "none") both are >> valid. But there is no order in which the PEs are selected. I would suggest >> to leave the "pe_all" out of the game (or name it different to address it >> directly). >> >> -- Reuti >> >> >>> Consider that @hgrp1 is on IB switch 1, @hgrp2 is on IB switch 2, and >>> @hgrp3 is on IB switch 3; that @allHosts = @hgrp1,@hgrp2,@hgrp3; that >>> switches 1-3 are connected to a 4th switch. >>> >>> Then, we do something fun like >>> >>> -pe pe* <#slots> >>> >>> to get a poor man's best-effort IB topology scheduling. Would it work? >>> >>> -Brian >>> >>> Brian Smith >>> Sr. System Administrator >>> Research Computing, University of South Florida >>> 4202 E. Fowler Ave. SVC4010 >>> Office Phone: +1 813 974-1467 >>> Organization URL: http://rc.usf.edu >>> _______________________________________________ >>> users mailing list >>> users@gridengine.org >>> https://gridengine.org/mailman/listinfo/users >> > > _______________________________________________ > users mailing list > users@gridengine.org > https://gridengine.org/mailman/listinfo/users _______________________________________________ users mailing list users@gridengine.org https://gridengine.org/mailman/listinfo/users
