Newsgroups: comp.parallel.mpi,comp.parallel.pvm,comp.parallel
From: "Joseph R. Matarese" <matarese@mit.edu>
Subject: MPI-based PVM implementation?
Organization: MIT Earth Resources Laboratory
Date: 5 Aug 1996 17:07:18 GMT
Message-ID: <32060373.2781E494@mit.edu>

Folks -

I've been digging around quite a bit and haven't unearthed any
information regarding an MPI-based PVM implementation.  In the book,
"PVM", published by MIT press a while back, there's a sentence
mentioning that PVM can be layered over MPI to provide all of those
useful features of PVM on top of a standard message passing interface
(MPI).  There's also brief mention on the PVM pages of a port to MPI.

My question: is anyone actually working on this?  

These ongoing "benchmarks" between MPI and PVM strike me as a
phenomenally ridiculous waste of effort.  That HPCC sponsor agencies are
funding people to do yet another benchmark seems like a complete waste
of taxpayers' money.  Why not unify and get on with useful work?
Following unification the MPI people could keep plugging away on
low-level standards, and the PVM people could enhance PVM's high-level
functionality.  Academia and the national labs would reap significant
benefits, and the market for HPC vendors might actually become more
profitable.  Is it not that simple?  Or is it like the old cliche, "The
best thing about standards is that there are so many to choose from!"

joe

-- 
MIT ERL, E34-556, 42 Carleton St., Cambridge, MA 02142  (617)-253-3992


