Path: utzoo!utgpu!water!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!mailrus!ames!umd5!uflorida!gatech!hubcap!fpst
From: fpst@hubcap.clemson.edu (Steve Stevenson-Moderator)
Newsgroups: comp.parallel
Subject: Shared mem vs msg passing
Keywords: shared memory, message passing
Message-ID: <1801@hubcap.UUCP>
Date: 2 Jun 88 12:04:55 GMT
Sender: fpst@hubcap.UUCP
Lines: 31
Approved: parallel@hubcap.clemson.edu

Approved: parallel@hubcap.clemson.edu

Shannon Nelson 
>     1. Which architecture is easier to program for, and why;

	This is not a simple question since most folks do not have
enough experience on both.  Each has obvious advantages, but most
of the criteria are performance ones.  The operating system folks have
the most experience and they probably can't tell you either.

	Languages are a problem here too.  Read Greg Andrews article
in the Jan 88 TOPLAS.

>     2. Does #1 depend on the application, and why;

	Undoubtedly.  Example: 'cubes can be added to ( FPS T-series goes
from 8 to 4096 processors.)  If the system is programmed distributively,
then needs for memory (for e.g.) are easily met (Ah, hello Turing) by
your company's comptroller.  Not everything distributes easily - but
we don't have the experience yet.

>     3. Who makes the "best" machine of the two architectures, and why.

	I'll take the Fifth Amendment.  But I personally think that
this is not answerable without - pardon my math background - 
cost criteria for  "best".

Steve Stevenson                            fpst@hubcap.clemson.edu
(aka D. E. Stevenson),                     fpst@clemson.csnet
Department of Computer Science,            comp.parallel
Clemson University, Clemson, SC 29634-1906 (803)656-5880.mabell