Path: utzoo!attcan!uunet!cs.utexas.edu!tut.cis.ohio-state.edu!unmvax!ogccse!cvedc!nosun!fpssun.fps.com!celit!hutch
From: hutch@fps.com (Jim Hutchison)
Newsgroups: comp.arch
Subject: Chaotic motion in caches (was Re: flexible caches)
Keywords: chaos cache prediction
Message-ID: <640@celit.fps.com>
Date: 18 Sep 89 21:36:38 GMT
Sender: daemon@fps.com
Reply-To: hutch@fps.com (Jim Hutchison)
Organization: FPS Computing
Lines: 19
References:

<>
In some recent reading on chaotic systems I was intrigued by the possibility
of monitoring relaxation functions of computer systems.  In the passage I
found intriguing, the author was talking about water droplets (Note that
some of this was in a recent TV show on Chaos also).  The droplets would
neatly reform if the water was not forced to quickly through the dropper.
If forced out too quickly, the relaxation and reformation would become
unpredictable by standard methods.  The bright side of all this is that
by measuring it and plotting based on the deltas in the relaxation times
a "nice" curve appears.  This curve has a complex function.

Instead of properties of the physical system, properties of the computing
system could be examined.  The cache algorithm could play the equation
like a tape.  This has some problems with how the prediction data is stored,
and how it is used.  Since it is based on the execution path, it does not
fit well as bits in the instructions.

/*    Jim Hutchison   		{dcdwest,ucbvax}!ucsd!celerity!hutch  */
/*    Disclaimer:  I am not an official spokesman for FPS computing   */