Hi, Am 14.10.2013 um 19:22 schrieb Joseph Farran:
> We have several users using large job-arrays using 1-core per job array > element. > > We have a shared queue pointing to several 64-core nodes. > > With the above setup, each of our 64-core nodes ends up with 64 individual > jobs from > various users. This is normal and expected behavior. > > Is there a way to request a node such that when user A grabs say compute-1-1, > that no > other user can use that same node but allow user A to keep scheduling jobs to > compute-1-1 as long as user A has 1 or more cores in use on compute-1-1? It's possible to use an advance reservation with different amounts of core requests for the the reservation and the jobs. Something like: $ qrsub -pe smp 64 -d 72:00:00 $ qsub -ar <ar_id> -pe smp 1 -t 1-100 job.sh (But requesting a PE even with one slot is necessary to gain access to this AR.) -- Reuti > I tried using the complex resource: > > $ qconf -se compute-1-1 > hostname compute-1-1.local > load_scaling NONE > complex_values exclusive=true > > $ qconf -sc | grep exclusive > exclusive excl BOOL EXCL YES YES 0 > 1000 > > > And requesting it via qsub with: > > #$ -l exclusive=1 > > And it works, except that it only allows 1-core per user. What is the > trick to > setting up the consumable so that it will allow up to 64 cores and only from > the same user? > > Joseph > > _______________________________________________ > users mailing list > [email protected] > https://gridengine.org/mailman/listinfo/users _______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
