Newsgroups: comp.parallel.mpi,comp.parallel.pvm
From: ijb@somewhere (Ian Bush)
Subject: Re: Why explicit message passing??
Organization: Daresbury Laboratory
Date: 8 Apr 1998 13:25:58 GMT
Message-ID: <6gftt6$osq@mserv1.dl.ac.uk>

In article <vofpvit2c81.fsf@motherlode.llnl.gov>, Rob Neely <neely@motherlode.llnl.gov> writes:

SNIP !

>
>Many people in the MPI/PVM world would probably consider themselves in
>the field of "High Performance Computing" - so you've stepped into a
>forum where performance and scalability will almost always come before
>the latest unproven CS solutions, for better or worse.
>

I think that is the vital point. Though the latest wonderful CS 
solutions may be incredibly elegant and to many conceptually
much simpler than explicit message passing, if they don't perform
they ain't going to get used. This is further reinforced by
the fact that, at least here in the U.K. on the machines to
which I have access, you have to `pay' for the high performance
computing resources in some some sense, either by having a finite
quota of node hours available, or in real money. Ouch, the magic
word pay, and suddenly performance becomes even more important,
for a 10% improvement in performance becomes 10% more science
you can do this year, which means 10% more papers at the end of 
the year. And what is your quota of node hours dependent upon ?
You guessed it ... The amount and quzlity of Science you did
with your last allocation.

That said I also take Kamran's point that there are alternative
methods out there that are becoming more efficient, and traditionalist 
"Fortran + message passers" like myself should not close our eyes
to them. Who knows, 10 years hence they could be competitive. If
they do markedly ease programming distributed memory boxes, great !
( Ermmmm, no, actually not great, if everybody can code them maybe
I'll be out of a job :>) )

Finally I'd like to thank Kamran for staring a most interesting thread,

Ian

-- 

N.B.: Reply address munged. Please reply to I.J.Bush at
dl.ac.uk