Am 27.08.2013 um 09:40 schrieb Abhishek Bhattacharjee: > Hi Reuti, > > Thanks for your response. > > Is there no way to assign a particular number of a complex resource to the > queue and not to every queue instance. > > For example I want to allocate only one ncsim_simulator license to queue > regression.d, so that if one host uses up the license the other host in the > queue cannot run the second job until the first host completes the job and > releases the license.
So you have several queues in the system? In this case assign the complex on a global level and at the same time to the queue like you have right now. The tighter constrain will be used, hence the job requesting this resource can run only one at a time (due to the global assignment) and only in the regression.d (due to the assignment to the queue). In principle you could also use an RQS (i.e. limit queues regression.d to slots=1 [or jobs=1 in case you define such a complex]), but the complex definition is more straight forward IMO, RQS is more for fine tuning) -- Reuti > Abhishek > > -----Original Message----- > From: Reuti [mailto:[email protected]] > Sent: 27 August 2013 12:17 > To: [email protected] > Cc: [email protected] > Subject: Re: [gridengine users] Resource management issue using sge > > Hi, > > Am 27.08.2013 um 07:36 schrieb Abhishek Bhattacharjee: > >> I am trying to use SGE for managing regression jobs and licenses for my > project. >> >> I am also using flex-grid (using a loadsensor) to manage the flexlm > licenses. The qstat (qstat -F ncsim_simulator) report for one of my queues > is shown below. >> ncsim_simulator is defined as a complex. Only one ncsim_simulator >> resource is made available to queue regression.d (See screenshotof >> host configuration using qmon below) >> >> regression.d@telsrv11 BIP 0/0/4 5.43 linux-x64 > >> qc:ncsim_simulator=1 >> > ---------------------------------------------------------------------------- > ----- >> regression.d@telsrv43 BIP 0/0/4 0.06 linux-x64 > >> qc:ncsim_simulator=1 >> >> My problem is that when my regression job runs, two jobs get fired - one > on regression.d@telsrv11 and one on regression.d@telsrv43 even though I have > allocated only one resource of ncsim_simulator to the queue regression.d. > > In this case each queue instance gets the specified value of this resource. > > > A complex_value can be attached to several objects: > > 1) In case you have a resource only a limited number of times for the > complete cluster on a global level: > > qconf -me global > > 2) > > In case you have a resource only a limited number of times per exechost > (like memory): > > qconf -me <hostname> > > 3) > > In case you have a resource only a limited number of times per queue > instance (i.e. maybe even several times per exechost, mostly for boolean > values to keep special jobs in dedicated queues): > > qconf -mq <qname> > > -- Reuti > > >> Request help in understanding what I am missing in the configuration >> of the queue (or any other configuration) >> >> Thanks, >> Abhishek >> >> >> <Outlook.jpg> >> >> _______________________________________________ >> users mailing list >> [email protected] >> https://gridengine.org/mailman/listinfo/users > > _______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
