[Paraview] Paraview ( MPICH, OpenMPI or LAM/MPI )
kmorel at sandia.gov
Mon Jan 14 10:32:58 EST 2008
> I managed to install all the requirements for Paraview, I have
> installed MPICH and LAM/MPI and am using 2 node machines ( a beawoulf
> cluster ). Anyway, just a curious thought, and some questions that i
> would really like to know as I will be shifting this into a 32 node
> a) I have configured LAM/MPI, but am having problem with the
> MPI_INCLUDE_PATH and MPI_LIBRARY not found. I have installed it into
> my home directory, and I am not root.
What problem? You should be able to use the ccmake UI to point to the
include directory and the libmpi.a or libmpich.a (or equivalent .so
> b) Should I just switch to OpenMPI, also.. ( how does it make a
> difference to the installation of Paraview ), Any notes or tutorial I
> can read, i have tried searching the net but there is not much.
That's up to you. Our cluster administrators moved to OpenMPI because
it is supposed to be faster that MPICH, but I have never done any
performance tests on ParaView to know one way or another.
ParaView compiles with OpenMPI just like any other MPI implementation.
The only caveat is that you need to use the libmpi_cxx.so library in
addition libmpi.so. The reason is that even though ParaView does not
use any of the C++ MPI bindings, the OpenMPI version of mpi.h
automatically includes some of them. It's easiest to just add the
libmpi.so to the MPI_EXTRA_LIBRARY CMake variable.
> c) do I really need an extra mpi library or is MPICH good enough.
Not unless you are getting link errors. One of our previous clusters
used Myrinet, and the version of MPICH that uses it required linking to
the gm library. But if ParaView compiles with your installation of
MPICH without any linker errors, then you don't need to put anything for
the MPI_EXTRA_LIBRARY variable.
More information about the ParaView