Am 13.03.2011 um 20:44 schrieb Tal Hazan: > I'm using sge for simple job queuing, not MP/threads.
Why do your jobs then use 10, 5 and 3 slots? What's the reason behind it, when it' only a serial job? > What is the best solution for my setup to use complex configuration? From > your explanation I understand that requesting 10 slots * 1 requested license > = 10 licenses that I don't have. As I wrote: You can use "consumable JOB" if it shouldn't be multiplied (`man complex`). > That means I will need to go back to slot=core setup? To avoid overloading, it's a usual setup to have slots=core per queue and limit it lateron in case you have more than one queue per exechost. -- Reuti > Tal. > > -----Original Message----- > From: Reuti [mailto:[email protected]] > Sent: Sunday, March 13, 2011 9:42 PM > To: Tal Hazan > Cc: [email protected] Users > Subject: Re: [gridengine users] Complex configuration messes up pe_slots > > Am 13.03.2011 um 20:33 schrieb Tal Hazan: > >> Hi Reuti, >> >> Thanks for the quick reply. >> >> What I tried to do with my setup is to assign each host 40 slots, we have >> jobs that uses 10, 5 and 3 slots. >> >> It tried to configure qrsh to use -l slots, but it requested -pe slots >> instead. >> >> If there's a way to assign slots to job without using PE I will be >> glad to hear ! :) > > No, it's the way to go. When your application uses forks / threads / OpenMP, > I would name the PE e.g. "fork" / "thread" / "openmp". > > It just looks confusing to me, as it's no longer a plain serial job. But it's > personal taste how to name it I must admit. > > -- Reuti > > >> >> Thanks, >> Tal. >> >> -----Original Message----- >> From: Reuti [mailto:[email protected]] >> Sent: Sunday, March 13, 2011 9:08 PM >> To: Tal Hazan >> Subject: Re: [gridengine users] Complex configuration messes up >> pe_slots >> >> Hi, >> >> Am 13.03.2011 um 17:18 schrieb Tal Hazan: >> >>> I encountered a strange issue with sge6.2u5 : >>> >>> I've added new variable, 'vcsruntime_net' with Y/Y/0/0 settings and added >>> it to global host settings with value of 8. >> >> here you use 8 for "vcsruntime_net". >> >> >>> Using the following command puts the job in 'qw' mode, scheduling >>> information shows: cannot run in PE "serial" because it only offers 8 slots. >>> >>> root$ qrsh -p 0 -now no -q all.q -pe serial 10 -l vcsruntime_net=1 >>> -display nx2:2001.0 -cwd -V -verbose -N simv simv >> >> And here you request 10 (as it's multiplied by the slot count). Hence it >> can't run, as you have only 8 globally available. You can use "consumable >> JOB" if it shouldn't be multiplied. >> >> Can you explain the intention of your setup - I'm puzzled that a PE is named >> "serial". >> >> -- Reuti >> >> >>> it looks like -pe serial 8 considers vcsruntime_net as the slots setting. >>> >>> qstat -f: >>> queuename qtype resv/used/tot. load_avg arch >>> states >>> --------------------------------------------------------------------------------- >>> all.q@rtlsrv03 BIP 0/30/40 3.07 lx24-amd64 >>> --------------------------------------------------------------------------------- >>> emergency@rtlsrv01 BIP 0/0/1 0.02 lx24-amd64 >>> --------------------------------------------------------------------------------- >>> emergency@rtlsrv02 BIP 0/0/1 1.00 lx24-amd64 >>> --------------------------------------------------------------------------------- >>> emergency@rtlsrv03 BIP 0/0/1 3.07 lx24-amd64 >>> --------------------------------------------------------------------------------- >>> all.q@rtlsrv02 BIP 0/0/40 1.00 lx24-amd64 >>> --------------------------------------------------------------------------------- >>> all.q@rtlsrv01 BIP 0/0/40 0.02 lx24-amd64 >>> >>> ##################################################################### >>> # >>> ###### >>> - PENDING JOBS - PENDING JOBS - PENDING JOBS - PENDING JOBS - PENDING >>> JOBS >>> ############################################################################ >>> 143719 0.40756 simv root qw 03/13/2011 16:32:33 10 >>> >>> >>> Qconf -sp serial: >>> pe_name serial >>> slots 120 >>> user_lists arusers deadlineusers defaultdepartment rtl verification >>> xuser_lists NONE >>> start_proc_args /bin/true >>> stop_proc_args /bin/true >>> allocation_rule $pe_slots >>> control_slaves FALSE >>> job_is_first_task TRUE >>> urgency_slots min >>> accounting_summary TRUE >>> >>> Thanks, >>> tal. >>> _______________________________________________ >>> users mailing list >>> [email protected] >>> https://gridengine.org/mailman/listinfo/users >> >> > > _______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
