[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Yes, they are special, but...

    Date: Fri, 9 Jun 89 12:22 EDT
    From: barmar@Think.COM (Barry Margolin)

	Date: Fri, 9 Jun 89 09:16 CDT
	From: ai.gooch@mcc.com (William D. Gooch)

	    ...."highly specialized hardware"....

	I've heard this idea bandied about enough that I suppose I should
	understand it by now, but - sorry if I'm being dense - I still don't
	know what it means.  The term "specialized," usually stated as in
	opposition to "general-purpose," implies that there are things one
	cannot do on a Symbolics machine which other hardware does support.
	If you don't mind my asking, what are those things?  

    It's not that Symbolics machines can't do things that other machines
    can, it's that they don't do many common things well.  Symbolics
    machines are optimized for Lisp processing, and perform poorly at just
    about everything else.  I/O performance is abysmal.  It doesn't matter
    how fast you can do generic arithmetic if the machine grinds to a crawl
    when it pages.  They claim that they run C code at about the same speed
    as a Vax 11/750; meanwhile, Sun-4 users have desktop machines that get
    at least 5 times that performance for a fraction of the cost.  The Suns
    even beat the pants off them in Lisp processing; our code runs in both
    Genera and Lucid, and compiles and loads several times faster in Lucid.

I believe that the only reason your code compiles and loads several
times faster is because of you earlier observation about I/O
performance.  Specifically disk I/O.  We've used Lucid on the Sun and
originally thought Lisp processing to be faster on the Sun but after a
while we discovered:

1) By the time you load it up with enough memory and disk to make lisp
run decently on a Sun you've just spent enough money to buy a 3670.

2) The sun runs circles around symbolics if your programs are
relatively small, don't cons a lot (excerise the gc very much), and
don't run for long periods of time (over an hour or so).

3) Circuit simulations ran about 20% faster on our 36xx machines than on
our suns.  These simulations were done using a homegrown version of
Spice written in Fortran which does scads of matrix-math computations.
Fortran runs faster on Symbolics than on Sun?  Perhaps there is a
reasonable explanation for this.

But still, this is all apple<->orange type comparisons.  However I will
concede that the oranges are catching up quickly (at least on the
hardware end).

-Mark Alexander

    Note that in the above comparisons, the Symbolics machines are 3600
    series, not Ivory machines.  I don't know whether they've improved the
    I/O performance in the XL400.

    I love my Lispm, and won't give it up.  It's a great hacker workstation,
    and I love poking around in Genera.  But to be called "general purpose",
    a machine has to perform reasonably at lots of different tasks, and
    these machines don't.