Hi Jeff, Jeff Squyres wrote:
You get the idea.
This is pretty much the position I reached after talking with other MPI developers and some ISVs. Bill's paper shows that it is technically possible, what is missing is the critical mass. We may have that today.
There's a slight performance penalty for the translation layer, but for those who want an MPI ABI, this might well be an acceptable price to pay.
This performance penalty would be negligeable for the current MPI operations on today's machines. This common external interface could be designed to do the translation faster.
This design has my vote. Patrick -- Patrick Geoffray Myricom, Inc. http://www.myri.com
