Path: utzoo!utgpu!news-server.csri.toronto.edu!bonnie.concordia.ca!uunet!csun!news
From: swalton@solaria.csun.edu (Stephen Walton)
Newsgroups: comp.sys.amiga.advocacy
Subject: Re: Commodore Research and Development.
Message-ID: <1991Jan16.190834.2106@csun.edu>
Date: 16 Jan 91 19:08:34 GMT
References: <17561@cbmvax.commodore.com> <242^HH|@rpi.edu> <17616@cbmvax.commodore.com> <1991Jan16.052537.13735@zorch.SF-Bay.ORG>
Sender: news@csun.edu (News Administrator)
Reply-To: swalton@solaria.csun.edu (Stephen Walton)
Organization: Cal State Northridge
Lines: 21
In-Reply-To: xanthian@zorch.SF-Bay.ORG (Kent Paul Dolan)

[Please note that Ken has the Followup-To line on the original message set to
'comp.sys.amgia.advocacy.']

A similar thrashing effect with virtual memory can be demonstrated
with a single program which does a Fast Fourier Transform of a large
array.  The standard FFT algorithm requires pairs of elements of the
array separated by 1 at pass 1, 2 at pass 2, 4 at pass 3, and on up to
N/2 at step log_2(N).  This causes horrible effects.  Once again, one
sees that the time for the FFT grows as expected as N*log_2(N) for N
smaller than the working set, then explodes to much larger values.

Before re-becoming an astronomer, I programmed hypercube architecture
computers, and the FFT was my specialty.  One of the big advantages
they have is lots of RAM:  an AMETEK (now defunct) 64 node system had
64 MB of real RAM and could do a 32 MB FFT several times faster than a
VAX with 8 or 16 MB.

So, VM is not The Solution To All Your Woes.
-------------------------------
Stephen Walton, Dept. of Physics & Astronomy, Cal State Univ. Northridge
   I am srw@csun.edu no matter WHAT the stupid From: line says!