[Paraview] need/share parallel 32-partitions data

[Berk Geveci]

> I tested data from two independent sources.
> 1-legacy
> 2-xml binary data
> both work for a 16-piece data set using 4, 8, and 16 processors.
> neither works for a 32-piece data set for 4, 8 or 16 processors. Therefore
> it seems that the number of pieces of data cannot exceed the maximum number
> of allowed processors which seems to be 16.

Nope. No such limitation exists. VTK_MPI_MAX_NUMPROCS is only used by
the testing framework to determine how many processes the tests should
use (the same with some of the other VTK_MPI_XXX flags). There is
something else that is causing the problem here. I would not use the
legacy formats by the way. They are not designed to support
distributed data. Out of curiosity, how do you create a 32 piece
legacy file? How are you creating a multi-piece VTK XML datasets?
There are multiple ways of doing it.

As I mentioned it before, the best way of testing parallel
functionality is to start with datasets/source that are known to work.
For polygonal models, I would use the sphere source and jack up the
resolution. For structured data, I would use the wavelet source. For
unstructured data, I would use the wavelet source and then
tetrahedralize it. All of these should work well with > 16 processors.

-berk



> The installation was from the source and not cvs. Thus cmake was used.
> I found this with google
>
> //Maximum number of processors available to run parallel applications.
> // (see /home/mig/installed/vtkCVS/VTK/CMakeLists.txt for more
> // info.)
> VTK_MPI_MAX_NUMPROCS:STRING=1
>
> //Flag used by mpi to specify the number of processes, the next
> // option will be the number of processes. (see
> /home/mig/installed/vtkCVS/VTK/CMakeLists.txt
> // for more info.)
> VTK_MPI_NUMPROC_FLAG:STRING=-np
>
> But don't know to do with it once the installation has been made.
> Stephen
>
>> -berk
>>
>> On Fri, Aug 8, 2008 at 4:21 PM, Berk Geveci <berk.geveci at kitware.com>
>> wrote:
>>
>>>
>>> Hi Stephen,
>>>
>>> Does this happen at startup or when you load the data? Did you try to
>>> create a simple source such as the sphere source? We have run ParaView
>>> on more nodes than 32 so it has been done before...
>>>
>>> -berk
>>>
>>> On Tue, Aug 5, 2008 at 11:55 AM, Stephen Wornom
>>> <stephen.wornom at sophia.inria.fr> wrote:
>>>
>>>>
>>>> PV works fine on our cluster for the test case using max 8 processors
>>>> and
>>>> 16-partritions but gives this error with 32-partitions
>>>>
>>>>
>>>>
>>>> ERROR: In
>>>>
>>>> /tmp/rpmbuild/BUILD/paraview-3.0.2/Servers/Common/vtkProcessModuleConnectionManager.cxx,
>>>> line 180
>>>> vtkProcessModuleConnectionManager (0xa653820): Failed to set up server
>>>> socket.
>>>>
>>>> mpiexec: Warning: task 0 died with signal 11 (Segmentation fault).
>>>> mpiexec: Warning: tasks 1-7 exited with status 1.
>>>>
>>>> Question is:
>>>> 1-my data for 32-partitions is wrong or
>>>> 2-the paraview-mpi installation is not correct
>>>>
>>>> If someone has run 32 processors successfully, a solution to find the
>>>> our
>>>> problem is:
>>>> 1-I give you our data to test on your installation or
>>>> 2-You give me your data for say 32 processors and I test it on our
>>>> system.
>>>>
>>>> Stephen
>>>>
>>>> _______________________________________________
>>>> 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
>>
>
>