[Paraview] ParaView & MPI Question(s)

[Berk Geveci]

Troy,

ParaView requires very basic MPI support, all in MPI 1 standard. If
ParaView failed to compile with an MPI implementation, I am guessing
this is not because ParaView is using an MPI feature that is not
available. So far, I compiled ParaView with mpich (1 and 2), lampi,
lam/mpi, openmpi, vendor specific mpis etc... You can find almost all
MPI calls made by ParaView in VTK/Parallel/vtkMPICommunicator.cxx in
the source. There are a few more in vtkMPIController.cxx and
vtkMPIGroup.cxx. It is possible to compile hdf5 with MPI/IO support
but I don't think it is on by default. If it is, it shouldn't be hard
to by-pass it.

The MPI related parameters that are available in the configuration,
such as maximum number of nodes, have nothing to do with how paraview
is run by the user, rather they are used by the standard test system
(ctest). I would ignore them unless you are running tests or setting
up a dashboard.

-berk


On 6/25/07, Troy Telford <ttelford.groups at gmail.com> wrote:
> I hope this is a quick question:
> What requirements are there for using ParaView with MPI?  (ie. MPI 2, etc)
>
> I've been successful in compiling ParaView (and VTK) using a couple MPI
> implementations (Open MPI, for example), but not others, such as MVAPICH2.
>
> Unfortunately, I don't get the luxury of saying "Use this MPI" to my users, so
> I'm mainly interested in eliminating MPI implementations which don't have the
> features required by ParaView.
>
> Also -- I noticed that you can configure the command used to execute 'mpirun'
> using CMake.  (Things like the max number of nodes, etc).  Is the node count
> something that is determined only at compile time, or is it configurable
> afterwards?
> --
> Troy Telford
> _______________________________________________
> ParaView mailing list
> ParaView at paraview.org
> http://www.paraview.org/mailman/listinfo/paraview
>