1Re: DRAFT Agenda for SLUG/SMBX Meeting of 15 December0

[David Gadbois <gadbois%sygmund.cgs.utexas.edu@RELAY.CS.NET>]

    Date: Mon, 18 Dec 89 17:21 EST
    From: miller@CS.ROCHESTER.EDU (Brad Miller)

1        Date: Mon, 18 Dec 89 15:44 EST
        From: Barry Margolin <barmar@Think.COM>

0   [...]

1        We asked them about this.  Their answer is that it's very hard to make
        money as a software company.  I believe them; Lotus and Microsoft are
        exceptions.

    But Symbolics has also made it clear that they can't compete with SUN, et.
    al. in making low cost workstations. Plus it's becoming pretty clear that
    these workstations are becoming powerful enough to run real lisp
    environments without special hardware.... if nothing else they just use
    their speed to simulate. IMHO, it seems like Symbolics' only viable option
    is to get their environment running on these machines since a vast majority
    of Lisp users can't justify to their management a 30-40k machine, when
    they're used to dealing with 5-10k workstations. Particularly when the
    hardware performance is no better and often significantly worse. Let's face
    it, a lot of good hardware engineers work for Sun and DEC et al. and
    Symbolics can't afford to keep up.

0The argument that Lisp machines are dead, put forth by John Mashey and
other comp.arch denizens, goes like this:  ``Commodity'' architectures
will beat out any special-purpose ones because they have the ecomomies
of scale derived from a large market to keep on the cutting edge of
technology.  Special-purpose architectures will lag behind on the
technology curve except for the money-is-no-object markets like
supercomputing (and even those may be shrinking).  So, even though your
SPARCstation doesn't have hardware support for run-time type dispatch,
multiple stacks, forwarding pointers, garbage collection assitance in
memory management, and the like, you can do it in software, run rings
around a (comparably priced) Lisp machine, and have Unix, too.

There are several problems with this argument, of course.  The benchmark
data supporting it are based on programs that don't reflect actual
applications; they don't take into account the support for serious
non-locality of data and other phenomena significant for Lisp (but,
then, neither does the MacIvory.)  And, most important, they don't take
into account the software that runs only on Lisp machines.

I don't think anyone buys a Symbolics machine or board these days just
because they want to run Lisp; rather, they want to run Genera and have
to buy the hardware to get it.

    [...]

    Date: Mon, 18 Dec 89 17:15:22 -0600
    From: forbus@cs.uiuc.edu (Kenneth Forbus)

    [...]

    The last place where I'm seeing any advantage for Symbolics systems is
    (a) debugger and (b) "endurance" of the envirionment, supporting
    multiple-day computations without a hitch.  Advantage (a) is almost
    gone, with the new version of Lucid I'm testing.  Advantage (b)
    similarly.  At university prices, I can buy a new workstation, which
    will perform as well or better, for the price I would pay for full
    maintenance on two Symbolics machines.

    Given what I am hearing, I'm afraid my only option is to start
    scraping my Symbolics machines.  Use them until they break, and let
    them die with dignity.  Sigh.

This has been the situation at the UT CS department.  The Symbolics
machines are off maintenance and are starting to die off as the disks
crash and the monitors fade away (the two most common failure modes).
The reasons:  the cost of hardware maintenace is far too high for the
old machines, it's a pain to manage another incompatible system, and
there is little incentive to buy Ivories now that standard (i.e., Unix)
platforms have reasonable Lisp environments that are faster and cost far
less.

I won't deny that Genera still beats any other environment I've seen,
but I think it's perceived as an luxury, and most of the fancy features
(dynamic windows, etc) that make it worthwhile are totally non-portable
to the machines that everyone else has (like Unix boxes).  Maybe CLIM
will help on this front.

Word on the street is that some folks are coming up with environments
that rival Genera, are written in C++, and run under Unix.  It may be
just talk, but the time sure is ripe.  I wonder how Symbolics will
compete a $10,000 software package with a $20,000 hardware package.

This note has been pretty pessimistic.  I just hope the Symbolics will
do the things it needs to do to keep up in tomorrow's market so that I
won't have to deal with the absurdities of Unix.


--David Gadbois