[GE users] scheduler runs performance sge6.0u6 vs. sge6.0u7

Nicolas Joly njoly at pasteur.fr
Tue Dec 13 09:36:01 GMT 2005

On Tue, Dec 13, 2005 at 09:23:47AM +0100, Stephan Grell - Sun Germany - SSG - Software Engineer wrote:
> Nicolas Joly wrote On 12/12/05 21:12,:
> >On Mon, Dec 12, 2005 at 08:43:29PM +0100, Reuti wrote:
> >  
> >
> >>Hi,
> >>
> >>this is a really short schedule_interval. Did you try to set it to a  
> >>minute or so and adjust the value of flush_submit_sec to one or two  
> >>seconds instead (seems you are looking for an immediate scheduling)?
> >>    
> >>
> >
> >I thought i already tested this configuration. Just to be sure, i did
> >it again ... and it works as expected.
> >
> >For now, we only have interactive jobs on this cluster (but i suspect
> >that in a few weeks we'll have to support some batch ...), and
> >immediate like scheduling is what we are indeed looking for.
> >
> >Thanks a lot.
> >
> Did the issue go away?

Yes. With schedule_interval set to 00:01:00 and flush_submit_sec=1, i
got the expected behaviour.

> Based on your profiling data, I would think that it does not make a
> difference. The scheduler spend its time in the communication with
> the qmaster and not processing anything.

I don't really understand why it takes so long ... The qmaster &
scheduler are located on a 2 CPUs opteron machine. And, last time i
checked, this machine load was very small.

> |P|PROF: send orders and cleanup took: 3.090 (u 0.010,s 0.000) s
> Your changes in the flush-settings should not change
> anything. However, the setup Reuti suggested is the one we
> recommend.

That's why i asked for help. I already tried tried the flush
parameters but got no difference. It was probably with
schedule_interval set to 1 second ...


Nicolas Joly

Biological Software and Databanks.
Institut Pasteur, Paris.

To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net

More information about the gridengine-users mailing list