1Re: Yes, they are special, but...0

[Chris Lindblad <cjl%life.ai.mit.edu@RELAY.CS.NET>]

    Date: Mon, 19 Jun 89 11:44 EDT
    From: miller@cs.rochester.edu (Brad Miller)

1	Date: Fri, 9 Jun 89 14:04 PDT
	From: pwtc!hamscher@labrea.stanford.edu (Walter Hamscher)

	Right, let's cut the doubletalk.

	Symbolics machines are the general-purpose machines because they
	deal with symbols, structured objects, and numbers equally well.

	It's all those other guys that are specialized because they only
	do arithmetic well and take relative performance hits to do the rest.

    Then why is Latex 20x slower (or worse) than a unix box? Genera is doing
    *something* poorly.... maybe it's just the Pascal compiler. But that just
    point up the issue of trying to compete with boxes where there are thousands
    of programmers generating applications/languages/etc. vs. one in which there
    are at most a couple hundred. So "general purpose" here means something like
    providing an environment with a lot of tools that work pretty well, vs. one
    in which LISP is terrific but other things, well... I'm getting tired of
    Zmail bugs, how about you?

0TeX compiled in Pascal performs similarly on the Symbolics 3650 running Genera
7.2 and the Digital VAX 11/750 running Berkeley UNIX 4.2.  This is as
expected, as the two machines are built out of the same technology (CMOS gate
arrays), have close to the same cycle time, and have lousy Pascal compilers.

TeX, translated to C, compiled in a optimizing C compiler, and running on a 20
MHZ CMOS VLSI microprocessor optimized for C is much faster than either of
these two machines.

I suspect TeX, translated to lisp, compiled in an optimizing Lisp compiler and
running on a 20MHZ microprocessor optimized for Lisp would run just as well as
it does on a UNIX box.

The only problem is that there's lots of people doing stuff like writing
optimizing C compilers and good pascal-to-C translators and building fast CMOS
VLSI chips optimized for C, while relatively few are doing the same for Lisp.

This is pretty much what the terms "out of the mainstream" and "not general
purpose" mean to me, and frankly I don't expect things to change for a while.
So let's not worry about it and use lisp machines for what they're good at:
Developing big systems.