[Mauiusers] blocked jobs never run
William G Wichser
bill at Princeton.EDU
Wed Nov 24 18:26:57 MST 2010
They are given higher priorirty but because of the fact that they cannot fit, they go into mthe blocked state and never get scheduled. I given wide jobs much higher weights. But again, because there are not enough nodes ever available at one time, they sit blocked with less wide jobs keep getting scheduled due only to available cores.
----- Original Message -----
From: Roy Dragseth <roy.dragseth at cc.uit.no>
Date: Wednesday, November 24, 2010 5:56 pm
Subject: Re: [Mauiusers] blocked jobs never run
To: mauiusers at supercluster.org
> On Wednesday, November 24, 2010 22:51:27 Bill Wichser wrote:
> > I seem to be having deja vu all over again.
> > The environment is torque/maui. There are a number of classes
> (queues)> set up, each with a limit for MAXPROCS. Lets take a
> class "A" and set a
> > maxprocs limit of 100. The queue is full and new jobs block as
> they now
> > exceed this limit.
> > So far so good.
> > Further, there is a 50 core job in there waiting with what would
> be the
> > highest priority if it were in the idle state. But it remains
> blocked.> Other jobs are also blocked by only want 8 cores. As
> jobs complete,
> > lets say they are 10 core and 8 core jobs, only these blocked 8 core
> > jobs ever get scheduled. The 50 core job always remains blocked
> because> resources are never available.
> > Yes one could use the JOBAGGREGATION time to try and wait for
> more to
> > finish in a timely manner but when this takes maybe 16 hours, it
> just> isn't feasible. Basically I have a blocked job which gets
> starved and
> > no way to get it scheduled with the current mix of jobs coming in
> and> moving through without manually disabling the "started" in
> qmgr or
> > placing jobs on hold until resources become available.
> > Is there some way to handle this that I just don't know or have
> figured> out? I hope that I'm just missing something simple.
> > Thanks. And for all the USofA folks, Happy Thanksgiving.
> Have you looked at the prioritization? Are big jobs given higher
> that small jobs? Have you allowed the highest priority jobs to
> reservations on nodes?
> mauiusers mailing list
> mauiusers at supercluster.org
More information about the mauiusers