[Paraview] Parallel Paraview

[David E DeMarle]

Slightly less briefly than what Ken said:

In parallel, the end of the pipeline on each processor is told to
process piece i out of j. Where i is the local processor's rank and j
is the number of processors. See
vtkStreamingDemandDrivenPipeline::SetUpdatePiece() and
SetNumberOfPieces(). That information flows back through the pipeline
in the RequestUpdateExtent pass. Here each filter along the pipeline
looks as what it is asked to produce, and tells its inputs what they
need to produce for it to do that. In your case if the algorithm
requires the whole domain then filter is telling everything upstream
to give it everything, and parallelism is only effective AFTER your
filter. If the algorithm can be changed so that each processor needs
only a portion of the domain (perhaps using ghostlevels to extent the
boundaries a bit) then you will get better scalability. After
RequestUpdateExtent, the RequestData pass happens in which each filter
executes in turn. In this pass the filter's RequestData() method uses
the input it is given to fill out the previously agreed extent and
passes it onward.

On Wed, Mar 4, 2009 at 10:26 AM, Natalie Happenhofer
<nataliehapp at hotmail.com> wrote:
> Hi!
> I´m not sure if I understood well how Paraview is working in parallel mode:
> I have started Paraview in client/server mode, i.e. with
> mpirun -np 8 ./pvserver
> and then I started the client and connected.
>
> I´m not sure how this works with filters: I´ve written a few filters, and
> when I run Paraview in parallel, is the filter parallelized as well,
> automatically? That on every piece of data the filter is applied separately
> and necessary data of other pieces is send there by mpi? Or is the filter
> computed on a single node and I have to change something in order to make it
> work parallel?
> Applying the ProcessId-Filter on the output of my filter, it says that it
> was computed separately on every piece of dataset, I´m just wondering,
> because the filter uses the whole dataset to compute the desired output..
>
> thx for explaining,
> Natalie
>
>
> ________________________________
> Hay un nuevo Messenger que tienes que conocer... ¡Mira las novedades!
> _______________________________________________
> Powered by www.kitware.com
>
> Visit other Kitware open-source projects at
> http://www.kitware.com/opensource/opensource.html
>
> Please keep messages on-topic and check the ParaView Wiki at:
> http://paraview.org/Wiki/ParaView
>
> Follow this link to subscribe/unsubscribe:
> http://www.paraview.org/mailman/listinfo/paraview
>
>



-- 
David E DeMarle
Kitware, Inc.
R&D Engineer
28 Corporate Drive
Clifton Park, NY 12065-8662
Phone: 518-371-3971 x109