Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!cs.utexas.edu!csd4.milw.wisc.edu!uxc!uxc.cso.uiuc.edu!m.cs.uiuc.edu!p.cs.uiuc.edu!forbus
From: forbus@p.cs.uiuc.edu
Newsgroups: comp.lang.lisp
Subject: Re: long and deep?
Message-ID: <80300008@p.cs.uiuc.edu>
Date: 12 Jun 89 05:01:00 GMT
References: <47400023@uicsrd.csrd.uiuc.edu>
Lines: 17
Nf-ID: #R:uicsrd.csrd.uiuc.edu:47400023:p.cs.uiuc.edu:80300008:000:931
Nf-From: p.cs.uiuc.edu!forbus    Jun 12 00:01:00 1989


I know I've built lists with over 100,000 elements easily.  But probably nested
only 20 or 30 levels deep.  However, that is only stuff I produce.  Alot of
stuff is produced by programs, and I don't look at every piece of internal
structure for huge examples.

Given your return address, I assume you care about this because of trying for
parallel speedups.  Lists aren't the only concern.
Most of one's non-locality nightmares come about through lots of
backpointers in structs.  The size of those datastructures, taken together,
can be in the 10's of megabytes.

I've heard random claims about making Lisp run on vector-oriented machines that
are based on "pure" Lisp, or worse yet, pure Scheme.  Anyone who thinks they
have made signficant progress because they've sped up a side-effect free Lisp
is fooling themselves.  Real AI systems almost always have massive
datastructures that evolve over the course of the computation.