That would be too easy :-)  I'm guessing my best bet would be to set CPUs=32 
and a default ntasks of 2
and override it if need be (I'd never get my users to do this reliably)

Any idea where I'd set this?  slurm.conf does not appear to have an entry for it

________________________________
From: Chrysovalantis Paschoulas [[email protected]]
Sent: 27 November 2014 14:54
To: slurm-dev
Subject: [slurm-dev] Re: overcommit on independent jobs

Try the option "--overcommit" from srun.

Check also the man page. You can not exceed the MAX_TASKS_PER_NODE limit of 
your configuration.

Cheers,
Valantis


On 11/27/2014 01:19 PM, Michael Lush wrote:
>So set: "CPUs=32" and maybe your problems will be solved ;)

Thankyou for answering :-)

That would work...  but is  would raise the number of jobs per node for 
everythen ..  16 jobs/node works well for almost all the workload,  we just 
want 32 for a particular application.  We could just run everything else with 
--ntasks=2,  but I was wondering if there was a way to do  --ntasks=0.5

--
Michael Lush
Computer Officer

CR-UK Genetic Epidemiology Unit
University of Cambridge
Strangeways Research Laboratory
Worts Causeway, Cambridge CB1 8RN
Tel.:  +44 (0)1223 748643
Web: http://ccge.medschl.cam.ac.uk/
________________________________




------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------
Forschungszentrum Juelich GmbH
52425 Juelich
Sitz der Gesellschaft: Juelich
Eingetragen im Handelsregister des Amtsgerichts Dueren Nr. HR B 3498
Vorsitzender des Aufsichtsrats: MinDir Dr. Karl Eugen Huthmacher
Geschaeftsfuehrung: Prof. Dr.-Ing. Wolfgang Marquardt (Vorsitzender),
Karsten Beneke (stellv. Vorsitzender), Prof. Dr.-Ing. Harald Bolt,
Prof. Dr. Sebastian M. Schmidt
------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------


Reply via email to