Newsgroups: comp.parallel.mpi,comp.parallel.pvm
From: Alan Williams <abwill@inreach.com>
Reply-To: abwill@inreach.com
Subject: Re: Why explicit message passing??
Date: Wed, 08 Apr 1998 06:48:13 GMT
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Message-ID: <352B1E81.D6EB58C3@inreach.com>

Well, my apologies again to Kamran, for my initial message which I repeat I
didn't mean
to send. You know how you get too many windows open, and you start pressing
buttons to close some of them... oh never mind.

Kamran Karimi wrote:

> [...]
>  Explicit message passing belongs to the stone age of the computer science.
> It is analogous to using machine language, and writing one's own hardware
> drivers instead of relying on general purpose operating systems. Most other
> branches of computer science have long left those times, but some distributed
> programmers seems to enjoy living in the past.
> [...]

Nobody would dispute that having automatic message passing would be nice,
explicit message passing at the application level is a pain, and is where a lot
of
programming errors occur. It's also hard to debug. Unfortunately there still
isn't
any other way to achieve really good performance on more than a few processors
if your application isn't 'embarassingly parallel'. There are research projects
which
attempt to provide logically shared memory in a variety of transparent ways,
but it isn't a solved problem yet, at least not for the general case. For the
case of a
machine with lots of processors, (hundreds or thousands), message passing is
still
the "state of the art", for now.
So I think that phrases like "stone age" and "enjoy living in the past" can't be
applied
to explicit message passing just yet. However, I'll concede that those opinions
are
not "ludicrous" either.  :-)

Alan