Path: utzoo!attcan!uunet!lll-winken!lll-tis!ames!mailrus!ncar!gatech!hubcap!jec
From: jec@iuvax.cs.indiana.edu (James E. Conley)
Newsgroups: comp.parallel
Subject: Re: A Taxonomy of Parallel Architectures
Message-ID: <3916@hubcap.UUCP>
Date: 15 Dec 88 18:10:36 GMT
Sender: fpst@hubcap.UUCP
Lines: 34
Approved: parallel@hubcap.clemson.edu


	In my architecture class we covered some other models in addition
to the Flynn model.  In particular there is the Feng, Handler, and Lipovski
models.  

	My notes are fairly sketchy, but I believe the Feng model classifies
machines by word size and number of bit-slices.  The types of machine can be
divided into (WPBS [word parallel, bit-slice serial], WSBS, WSBP, and WPBP).
You can then plot how "parallel" machines are relative to each other.

	The Handler model is an ordered set:

	T(arch) = <KxK', DxD', WxW'>

	where K is the number of processors, K' is the number of them that
are pipe-lined, D is the number of ALUs, D' is the number of them that are 
pipe-lined, W is the word size, and W' is the pipe-line depth.  The connection
machine would be:

	T(CM) = <32k x 32k, 1 x 1, 16 x 2>

	The last model is the Lipovski model and it's kind of complex.  The
idea is that it describes computing power by comparing the area of "bits
processed" times "time".  The shape of the area is supposed to give an
idea of the type of processor.   A fast but not very parallel machine will
loop like a horizontal rectangle whereas a slow but highly parallel 
architecture will look like a tall rectangle.

	Take my explanation with a grain of salt, these are from class notes
rather than any source that I've checked out myself.  If anyone has the
original references of these models, I'd be interested in seeing them.

James Conley
Indiana University Computer Science
jec@iuvax.cs.indiana.edu