The main problem with MPI is the huge number of function in the API.
Even if we implement only the 1.0 standard we still have several
hundreds of functions around. Moreover, an MPI library is far from
being a simple self-sufficient library, it requires a way to start
and monitor processes, interact with the operating system and so on.
All in all we end up with a multi-hundreds KB library which in most
of the applications will be only used at 10%.
We investigated this possiblity few months ago, but in front of the
task of removing all unnecessary functions from Open MPI in order to
get something that can fit in the 256KB of memory on the SPU (and of
course still leave some empty room for the user) ... Moreover, most
of the Cell users we talked with, are not interested to have MPI
between the SPU. There is only one thing they're looking for,
removing the last unused SPU cycle from the pipeline !!! There is no
room for anything MPI-like at that level.
On Mar 22, 2007, at 12:30 PM, Marcus G. Daniels wrote:
> Has anyone investigated adding intra chip Cell EIB messaging to
> It seems like it ought to work. This paper seems pretty convincing:
> users mailing list