Python MPI

[Eugene.Leitl at lrz.uni-muenchen.de]

Karsten Petersen wrote:
>
> Well, that only describes how to use Python with C-modules using MPI.

But that covers 95% of what I need, so it's plenty.

> I found it even more interesting to call MPI functions directly from
> Python - thats possible with e.g. the MPI-module contained in

I'm going to use this for realtime visualization, so there's no 
way I can call a bunch of Python code if I'm to expect ~10 fps
refresh, where each frame takes many MBytes of memory to stream
through.

> ScientificPython 2.1.
>         http://starship.python.net/crew/hinsen/scientific.html
> The above link also describes, that you have to patch your Python
> executable.  But my tests with lam-mpi and mpich had shown that it is
> not really needed - they finish themself cleanly even if MPI_Finalize is
> not called.

Good to know.
 
> > Any MPI Python users here? I'll be interested in hearing your experiences.
> I used Python with the MPI-Module from Scientific-Python 2.1 during the
> last months to do my MPI-coding homework.  Currently I am trying to
> write a parallizing fractal-generator in Python using MPI, NumPy and
> PyGTK...

I'm going to write a cellular automata MPI package around Rolf Rabenseifner's
heat transfer MPI demo, and the VolPack voxel rendering library, to run on
my 8-node K7 Beowulf (I guess DDR memory, since this is about memory stream
voxels).