[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Condor-users] Submitting MPI JOb
> On Mon, Jun 23, 2008 at 2:49 AM, <txcom2003@xxxxxxxxxxxxxxxxxxx> wrote:
>> No, I submitted the second job when Schedd has release those claims and
>> the status of all machine was UNCLAIMED. So the condition is same as the
>> first job.
> So it is likely that your first submission occurred just after a
> negotiation cycle, and had to wait for a majority of the negotiation
> cycle. The second submission may have occurred right before the next
> negotiation cycle, and got matched immediately.
> To help address this, you should play with the settings:
> NEGOTIATOR_INTERVAL - This affects how often negotiation occurs, and
> can be shortened depending upon some specifics about your
> installation. If you set this way too short (say 10 seconds), and you
> have machines on different networks that take a while to start jobs,
> you might get some thrashing. If your machines are close together
> network wise, you can safely set this smaller. I've made this very
> small depending upon the environment's network speed. Setting this to
> 30 can work in some environments provided you
> NEGOTIATIOR_CYCLE_DELAY - This is a required delay between any two
> negotiation cycles, and can be set smaller than the default (20
> seconds I think). I've made this 3-5 seconds in situations where all
> the machines are on the same switch. Even on larger installations
> putting this at 10-15 seconds is generally acceptable.
> Both of these changes should be made on the Central Manager's
> configuration. Additionally, Condor 7.0 contains a number of
> matchmaking/parallel job improvements, so if you're using 6.8, you
> might consider upgrading. As with all negotiation cycle tweaking
> you'll want to watch the NegotiatorLog to ensure that machines that
> were matched don't get unmatched during the next cycle because these
> settings were too low.
Thanks it works now.
Are the NEGOTIATOR_INTERVAL and NEGOTIATIOR_CYCLE_DELAY also only work in
condor version 7 ?
And how to improve matchmaking/parallel job ? is that already in default
Thanks a lot
> Hope this helps,
> Jason A. Stowe
> cell: 607.227.9686
> main: 888.292.5320
> Cycle Computing, LLC
> Leader in Condor Grid Solutions
> Enterprise Condor Support and Management Tools
> Condor-users mailing list
> To unsubscribe, send a message to condor-users-request@xxxxxxxxxxx with a
> subject: Unsubscribe
> You can also unsubscribe by visiting
> The archives can be found at: