Newsgroups: comp.parallel
From: wrankin@ee.duke.edu (William T. Rankin)
Subject: Re: SMP vs. MPP
Organization: Duke University EE Dept.; Durham, NC
Date: Mon, 2 Jan 1995 19:04:53 GMT
Message-ID: <3e49dd$bgq@news.duke.edu>

In article <9412202130.AA18009@idaho.SSD.intel.com>, Timothy G. Mattson <tgm@SSD.intel.com> writes:

|> There have been some attempts to merge SMP and MPP.  KSR is
|> the most recnet example.  They were unable to pull it off,
|> however, and have consequently gone out of business.

Actually, KSR "pulled it off" remarkably well.  The KSR-1 was a
wonderful platform to program on when using the threads programming
paradigm.  I was disappointed by their failure as a company, since
their "allcache" architecture showed so much promise.

Their failure as a business was due to many reasons, but their
"allcache" architecture was not one of them.  Also, the KSR is best
described as a "Distributed Shared Memory Multiprocessor".  I feel 
that using "Symetric Multiprocessing" and "KSR" in the same sentence
is misleading.


|> MPP systems (by which I mean distributed memory systems)

(minor nit with me) I really hate it when people start interchaging
"MPP" and "Distributed Memory Processors".  Unfortunately, the term
"MPP" is becoming so generic these days.  Kinda like how "RISC" became
a generic label for any processor with an ISA smaller than that of a 
Vaxen! :-)

(climbing off soapbox now...)

|> are
|> potentially scalable to hundreds if not thousands of
|> processors (both Intel and Ncube have deployed systems with
|> over 1000 processors and I have heard of a 512 node systems
|> from Cray and IBM).

Cray Research has advertised 2048 node T3D systems.  I have
played around on a 512 node T3D, so I can verify that it does exist.

|> I prefer MPP systmes, however, out of fairness I should point
|> out the SMP systems can be much easier to use.  Hence, while
|> us research types continue to get the most out of MPP systems,
|> the SMP computers are the ones making it in the market (in terms
|> of forming viable companies with broad user bases).  So take
|> your pick -- both have their advantages and disadvantages.

Good point.  It's the "ease of use" versus "max performance/scalability"
argument.

|> --Tim


-- 
----                                /       __/    /    /
bill rankin                        /              /    /
wrankin@ee.duke.edu               ___  /    /    /    /
philosopher/coffee-drinker       /    /    /    /    /
molecules-R-us                  /    /    /    /    /
                             _______/  __/  __/  __/

"All I wanna do is have a little fun before I die."  - Sheryl Crow


