[Paraview] mpirun of paraview

[Randall Hand]

On our SGI system here i've noticed the same thing.. It doesn't seem to
impact paraview's performance, but no matter what's going on all the
processes are pegged at 100% utilization.

It'd kinda funny in a way to see the utilization actually *drop* when they
start doing real work.

On 4/18/06, John Patchett <patchett at lanl.gov> wrote:
>
> It's probably MPICH, we have also witnessed the CPU spinning using
> MVAPICH.  We wrote some code that found nodes with open receives show
> high usage.  This makes sense that process 0 shows normal CPU usage
> while the rest of the nodes in the job have open receives while waiting
> for process 0 to tell them to do something.  I believe the behavior
> causing the high CPU usage is called polling and you can build MVAPICH
> (as well as MPICH??) with polling turned off.  I understand this will
> simply increase your latency.
>
> -John
>
> On Apr 18, 2006, at 12:22 PM, Sean Ziegeler, Contractor wrote:
>
> >> I tried to run paraview which was compiled with mpich. I noticed that
> >> cpu usage on paraview data servers (pvserver) is always around 99%
> >> even though no data processing at all. Only the first pvdataserver's
> >> cpu usage was reasonable(where I started the mpirun command).
> >>
> >> Why did that happen? Does that mean paraview is not working properly
> >> on all paraivew data servers.
> >
> > That might be an MPI issue.  We use MVAPICH (a derivative of MPICH) and
> > it also seems to peg-out the processors for all but the first process.
> > This happens with any MPI run, not just Paraview.  I am guessing MPI is
> > doing some sort of spin-wait on those processes.
> >
> > Anyway, to see if it truly is MPICH, you can try running some other MPI
> > program that can idle for a certain amount of time and see if those
> > processes behave similarly.
> >
> > -Sean
> >
> > _______________________________________________
> > ParaView mailing list
> > ParaView at paraview.org
> > http://www.paraview.org/mailman/listinfo/paraview
>
> _______________________________________________
> ParaView mailing list
> ParaView at paraview.org
> http://www.paraview.org/mailman/listinfo/paraview
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://public.kitware.com/pipermail/paraview/attachments/20060418/50165611/attachment.htm