[GE users] Help: Scheduler Profiles

Reuti reuti at staff.uni-marburg.de
Thu Nov 29 11:48:45 GMT 2007


Am 29.11.2007 um 11:55 schrieb Lee Amy:

> 2007/11/28, Reuti <reuti at staff.uni-marburg.de>:
> Am 28.11.2007 um 15:11 schrieb openlinuxsource:
>
> > Reuti wrote:
> >> Hi,
> >>
> >> Am 28.11.2007 um 12:34 schrieb openlinuxsource:
> >>
> >>> I read SGE Installation Guide, and I notice that the Scheduler
> >>> Profiles part will choose a scheduler profile for tuning. And I
> >>> have a problem about it.
> >>>
> >>> When I install SGE, I choose "normal" profile. In fact, I just
> >>> use default option. However, I find that the other two options
> >>> might be important in schedule strategy. The "max" profile can
> >>> enable immediate scheduling. So I suppose that it means
> >>> dynamically, I may need to use this profile for my cluster. But I
> >>> have chosen "normal", can I change a profile to another profile?
> >>> Is it will influence resource scheduling obviously?
> >>
> >> you can set with "qconf -msconf":
> >>
> >> schedule_interval                 0:2:0
> >> flush_submit_sec                  4
> >> flush_finish_sec                  4
> >>
> >> -- Reuti
> >>
> >>  
> ---------------------------------------------------------------------
> >> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> >> For additional commands, e-mail: users- 
> help at gridengine.sunsource.net
> >>
> >>
> > Thank you for your reply. If I enable this in "qconf -msconf", I
> > need to set up functional or share tickets, right?
>
> No. The settings above are just for max scheduling. Nothing more is
> necessary.
>
> -- Reuti
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net
>
> Reuti,
>
> Thank you for help. I've changed settings with the command. But  
> when I submit jobs, I still can't see the dynamically scheduling  
> from QMON directly.
>
> Mu cluster has five nodes with 20 processors and I set up SGE can  
> use 19 processors. When I submit jobs I find that SGE can run 19  
> jobs, other jobs are pending. If I wanna run more jobs than  
> processors, what should I do? I know I run the jobs more than  
> processors only, SGE can use its own dynamically adjust.

If you have a large SMP machine, SGE can be set up to use the  
functional policy to adjust the nice values of the processes  
dynamically. This means of course to oversubscribe the machine. If  
you have just "slots = processes", you can't do much to speed them up.

With a clusters of machine with 4 cores this would only work inside  
each machine.

-- Reuti



More information about the gridengine-users mailing list