Thanks a lot Jonathan, that is exactly what I was suspecting. In my case the optimum is 1 node and one thread then. Best regards, Murat
-- Prof. Murat Yildizoglu [email protected] > Le 4 nov. 2015 à 12:57, Jonathan Passerat-Palmbach > <[email protected]> a écrit : > > Hi Murat, > > Don't think jobs when you're in OpenMOLE. > Think of what each of the tasks in the workflow needs to run correctly > and as fast as you want. > > So if each of your 100 experiment require 1 node and 4 threads, that's > how you want to define your PBS environment. > > Cheers > J. > >> On 04/11/15 11:38, Murat Yildizoglu wrote: >> Hi Romain, >> The zip idea seems good, I will check it. >> >> Concerning the PBS, what I would like to understand is the following: >> >> Imagine that I have 100 experiments coming out from the sampling of the >> parameters >> >> How many jobs, each with which demands from cluster resources, OM will >> launch in the following two cases : >> >> nodes=50, >> threads=2 >> >> or nodes=1 >> threads=1 >> >> My current settings were >> >> nodes=50, >> threads=4 >> >> Hoping that OM would launch 200 jobs in parallel on 50 nodes, but it >> seems to give these requirements to each job it generates, which slows >> of course their position in the queue. >> >> I hope that I am clear enough this time :-) >> >> By the way, I love the new OM. >> >> Best regards, >> >> Murat >> >> 2015-11-04 11:40 GMT+01:00 Romain Reuillon <[email protected] >> <mailto:[email protected]>>: >> >> I don't exactly understand, what is your PBS env definition in >> OpenMOLE ? >> >> >> Le 04/11/2015 11:17, Murat Yildizoglu a écrit : >>> Hi again, >>> >>> (I hope that someone will be able go indicate me a trick to save a >>> configuration index in the CSV file) >>> >>> I am testing our PBS based cluster. But, I have the impression >>> that I am not setting correctly the node and threadsd parameters. >>> For now I have asked 50 nodes with 4 threads for all my >>> experiments but when I check the generated jobs has this >>> parameters for each job. Since I cannot see exactly the content of >>> the job generated by OM, I am not sure that this is the correct >>> setting. >>> Since I am on a local cluster, probably setting node=1 and >>> thread=1 would give a quicker advancement on the jobs queue, and >>> quicker execution of the jobs, no? >>> >>> Best, >>> >>> Murat >>> >>> >>> -- >>> Prof. Murat Yildizoglu >>> >>> Université Montesquieu Bordeaux IV >>> GREThA (UMR CNRS 5113) >>> Avenue Léon Duguit >>> 33608 Pessac cedex >>> France >>> >>> Bureau : E-331 >>> >>> mail: yildi-at-u-bordeaux4.fr <http://yildi-at-u-bordeaux4.fr> >>> >>> web: yildizoglu.info <http://yildizoglu.info> >>> >>> >>> _______________________________________________ >>> OpenMOLE-users mailing list >>> [email protected] <mailto:[email protected]> >>> http://fedex.iscpif.fr/mailman/listinfo/openmole-users >> >> >> _______________________________________________ >> OpenMOLE-users mailing list >> [email protected] <mailto:[email protected]> >> http://fedex.iscpif.fr/mailman/listinfo/openmole-users >> >> >> >> >> -- >> Prof. Murat Yildizoglu >> >> Note: Please use the following address as such >> >> UNIVERSITE DE BORDEAUX >> GREThA (UMR CNRS 5113) >> MURAT YILDIZOGLU >> 16 AVENUE LEON DUGUIT >> CS 50057 >> 33608 PESSAC CEDEX >> FRANCE >> >> Bureau : E-331 >> >> mail: murat.yildizoglu at u-bordeaux.fr <http://u-bordeaux.fr> >> >> web: www.yildizoglu.fr <http://www.yildizoglu.fr> >> >> >> _______________________________________________ >> OpenMOLE-users mailing list >> [email protected] >> http://fedex.iscpif.fr/mailman/listinfo/openmole-users > > -- > Jonathan Passerat-Palmbach, PhD > Research Associate > Department of Computing > Imperial College London > > South Kensington Campus > Huxley Building - room 344 > 180 Queen's Gate > London SW7 2AZ > _______________________________________________ OpenMOLE-users mailing list [email protected] http://fedex.iscpif.fr/mailman/listinfo/openmole-users
