[Beowulf] ANN: MorphMPI project (in alpha stage)
Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.
Toon Knapen toon.knapen at fft.beFri Aug 18 06:44:42 PDT 2006
- Previous message: [Beowulf] Java vs C++ for interfacing to parallel library
- Next message: [Beowulf] [PCGRID07] Call for Papers for Workshop on Desktop Grids
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Dear all, Some time ago there have been multiple discussions on an MPI ABI. Finally we concluded that we would be able to accomplish the same goal by putting a small layer on top of MPI (MorphMPI) that would make abstraction of the underlying MPI implementation. As usual, the solution is to add an extra level of abstraction ;-). Applications compiled against MorphMPI would be able to be relinked to any MPI implementation without having to be recompiled. So finally I developed a small prototype of such a MorphMPI. Using this MorphMPI I was able to compile an application on an AMD-Opteron and launch it using mpich-ch_p4, mpichgm and the preinstalled mpich on a Cray XD1. To make sure it would not only work with several derivatives of mpich I also tested it successfully on an HP-UX using mpich-ch_p4 and HP-MPI (which is LAM based). I uploaded my prototype to http://sourceforge.net/projects/morphmpi and you can find a (far too short) description at http://morphmpi.sourceforge.net/. If there is any interest in the library I plan to make it more user-friendly and extend it to cover the full MPI-1.1 standard. toon
- Previous message: [Beowulf] Java vs C++ for interfacing to parallel library
- Next message: [Beowulf] [PCGRID07] Call for Papers for Workshop on Desktop Grids
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
