Path: utzoo!utgpu!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!cwjcc!mailrus!uflorida!gatech!hubcap!munnari!tasis.utas.oz.au!ben
From: munnari!tasis.utas.oz.au!ben@uunet.UU.NET (Ben Lian)
Newsgroups: comp.parallel
Subject: Re: A Taxonomy of Parallel Architectures
Message-ID: <3937@hubcap.UUCP>
Date: 19 Dec 88 13:18:47 GMT
Sender: fpst@hubcap.UUCP
Lines: 63
Approved: parallel@hubcap.clemson.edu

In article <3920@hubcap.UUCP> eugene@eos.arc.nasa.gov (Eugene Miya) writes:
>Additional taxonomies:
> [stuff deleted]
>Taxonomies have very little real value except to Washington funding bureaucrats
>(but perhaps I am too harsh?).


Now that's an interesting point!  What would you want a taxonomy for anyway?
Other than to say "if you show me yours, I'll show you mine".  No, only
joking.

Seriously, it would be handy to be able to describe the general characteristics
of a particular parallel machine's architecture without having to spew mouthfuls
of words containing terms/acronyms whose meanings vary depending on where you
come from, what you've read, etc.  Taxonomies such as Flynn's are now too old
and hence too general to describe the different classes of machines that have
appeared on the scene.  I'm no expert (hence my original net request) but it
also seems that there is also a lot of confusion between 'architecture' and
'programmming model'.  For example, what do YOU (indicating the plural) take
the following grab bag of terms to mean:

.  shared memory / local memory
.  processors in lock-step (systolic?) / processors asynchronous
.  data driven (i.e., dataflow) / demand driven (i.e., graph reduction)
.  data parallel
.  wavefront
.  tightly coupled
.  closely coupled
.  loosely coupled
.  distributed architecture
.  MIMD
.  SIMD
.  <add your own>

Anyway, enough said.  I don't want to prolong discussion on this topic.  I
suspect that it rears its ugly head from time to time.  There may never be
a completely satisfactory and universally acceptable way of describing parallel
architectures.  Could it be that the scene is changing so rapidly that any new
taxonomy invented is almost immediately out of date?  Someone is bound to
invent/design a new machine which can be almost, but not quite, described by
existing terms.  Hence, I think, Eugene's remark above.

Oh, BTW, Eugene Miya has threatened to 'u' comp.parallel.  Something to do
with his parallel computing bibliography.  Perhaps I can prevail on Eugene
to NOT take such a drastic action, because this newsgroup will almost
certainly miss his "gross generalizations", eh?


Ben Lian




---------------------------------------------------------------------
Benjamin Y H Lian             ACSnet: ben@tasis.utas.oz
Dept. of EE & CS              ARPA  : ben%tasis.utas.oz.au@uunet.uu.net
University of Tasmania        BITnet: munnari!tasis.utas.oz!ben@
GPO Box 252C                          uunet.uu.net  (I think)
Hobart, Tasmania 7001         UUCP  : {enea,hplabs,mcvax,uunet,ukc}!
Australia.                            munnari!tasis.utas.oz!ben

"For every problem, there exists a solution that is simple, elegant,
and wrong."  [Why do *I* keep finding those solutions?]