[Mauiusers] Re: Contents of mauiusers Digest, Vol 8, Issue 2

David Jackson jacksond at clusterresources.com
Sat Mar 5 15:18:24 MST 2005


Jerry, Angel,

  We are already adding the ability to allow configurable 'relative'
wallclock (ie JOBMAXOVERRUN=30%) overruns to Moab.  We should be able to
roll this into Maui as well.

Dave

On Thu, 2005-03-03 at 11:15 -0700, Jerry Smith wrote:
> 
> 
> mauiusers-request at supercluster.org wrote: 
> > 
> > 
> > Today's Topics:
> > 
> >    1. What happens when jobs go over their walltime? (Angel de Vicente)
> > 
> > 
> > ----------------------------------------------------------------------
> > 
> > Message: 1
> > Date: Thu, 3 Mar 2005 11:58:14 +0000
> > From: Angel de Vicente <angelv at iac.es>
> > Subject: [Mauiusers] What happens when jobs go over their walltime?
> > To: mauiusers at supercluster.org
> > Message-ID: <16934.64470.94675.214413 at guinda.iac.es>
> > Content-Type: text/plain; charset=us-ascii
> > 
> > Hi,
> > 
> > I've been looking at the parameters for Maui, and I see that there is one called
> > JOBMAXOVERRUN that almost does what I want... I would like to be able to cancel
> > a job if it exceeds its walltime by, say, 50%. Can this be done somehow? It
> > looks unfair to set JOBMAXOVERRUN to a fixed amount. For example 1 hour of
> > overrun for a 1 week job is almost nothing, but for a 20 minutes one is quite an
> > overrun.
> > 
> >   
> > And assuming the job is cancelled, what happens to it? In our current
> > installation of OpenPBS we have a script that goes around the nodes killing the
> > processes of cancelled jobs. Do we have to write something similar or does the
> > Maui/Torque combination provide something already?
> > 
> >   
> The parameter  WCVIOLATIONACTION  can either preempt the job (provided
> your application handles preemption) , or cancel it entirely.
> > An aside: I also see in the documentation some interesting parameters, like
> > ENFORCERESOURCELIMITS, SYSTEMDEFAULTJOBWALLTIME or SYSTEMMAXPROCSECONDPERJOB,
> > but neither "showconfig" or "showconfig -v" show them. How can I know their
> > values?
> > 
> >   
> Looking in the documentation, you can see the default values of each
> Parameter adjacent to the parameter, ie..
> ENFORCERESOURCELIMITS
> <BOOLEAN> 
> FALSE
> 
> As for defaults for queuing resources, we let the server (torque)
> handle that.
> 
> Good luck,
> 
> Jerry Smith
> ---------------------------------
> jdsmit at sandia.gov
> Infrastructure Computing Systems
> Sandia National Laboratories
> ---------------------------------
> > Thanks a lot,
> > Angel de Vicente
> >   
> _______________________________________________
> mauiusers mailing list
> mauiusers at supercluster.org
> http://supercluster.org/mailman/listinfo/mauiusers



More information about the mauiusers mailing list