Re: [Condor-users] Changing the time interval for matching

["Timothy St. Clair" <tstclair@xxxxxxxxxx>]

If you want faster negotiations in the 7.4 series it's best to not hit
the negotiator but instead increase the CLAIM_WORKLIFE to let the schedd
reuse a claim.  This increases throughput rather dramatically, as
decreasing the NEGOTIATOR_INTERVAL too low can have an undesired affect.

You will also want to craft your HIBERNATE and START expressions in such
a way to match your policies.

Cheers,
Tim

On Thu, 2011-02-17 at 13:57 +0000, Stephen McGough wrote:
> Dear Peter, David, All,
> 
> Many thanks for this. We'd already tried taking NEGOTIATOR_INTERVAL down (to 20 seconds) without getting the desired result. Though we've not played around with SCHEDD_INTERVAL and I'll let you know if it has the desired effect.
> 
> Cheers,
> 
> Steve..
> 
> 
> On 16/02/2011 16:49, "Peter Doherty" <doherty@xxxxxxxxxxxxxxxxxxx> wrote:
> 
> 
> 
> On Feb 15, 2011, at 16:49 , David Brodbeck wrote:
> 
> >
> >
> > On Mon, Feb 14, 2011 at 7:23 AM, Stephen McGough <stephen.mcgough@xxxxxxxxxxxxxxx
> > > wrote:
> > Dear All,
> >
> > We have successfully set up a cluster which uses a non Condor method
> > for sleeping Windows 7 computers and uses Rooster to wake up these
> > computers when jobs are waiting. However, we now have a race
> > condition. The University here have a tight requirement for shutting
> > down computers "out of hours" and if no Condor job starts within 5
> > minutes the computer will be powered down. We have also seen that it
> > can take Condor 5 minutes to match and start a job on a computer
> > once Rooster has woken it up. At the moment we're seeing a number of
> > jobs which wake a computer up, fail to start in 5 mins so the
> > computer goes to sleep, just to wake up another computer. All jobs
> > now run but it would be good to remove these unwanted wake-ups.
> >
> > To do this we would like to reduce the amount of time Condor takes
> > to match (we're trying to extend the time interval before a computer
> > sleeps too). The START expression evaluates directly to true "out of
> > hours".
> >
> > You might try tweaking NEGOTIATOR_INTERVAL.  The default is 300
> > seconds before 7.4.0, 60 seconds after.  At our site we set it to 30
> > to reduce the amount of time jobs would spend sitting in the queue;
> > we have a small cluster so the extra load on the central manager has
> > not been a problem (or even noticeable, really.)
> >
> Also look at the SCHEDD_INTERVAL  You can decrease that as well,
> although a SUBMIT event should trigger a new SCHEDD cycle to start,
> unless you've tweaked that setting.
> 
> -Peter
> _______________________________________________
> Condor-users mailing list
> To unsubscribe, send a message to condor-users-request@xxxxxxxxxxx with a
> subject: Unsubscribe
> You can also unsubscribe by visiting
> https://lists.cs.wisc.edu/mailman/listinfo/condor-users
> 
> The archives can be found at:
> https://lists.cs.wisc.edu/archive/condor-users/
> 
> _______________________________________________
> Condor-users mailing list
> To unsubscribe, send a message to condor-users-request@xxxxxxxxxxx with a
> subject: Unsubscribe
> You can also unsubscribe by visiting
> https://lists.cs.wisc.edu/mailman/listinfo/condor-users
> 
> The archives can be found at:
> https://lists.cs.wisc.edu/archive/condor-users/