May Symbolics thrive...

[aboulang@BBN.COM]

   Date: Tue, 5 Jan 93 19:51:58 EST
   From: qobi@unagi.cis.upenn.edu
   Posted-Date: Tue, 5 Jan 93 19:51:58 EST
   Reply-To: Qobi@central.cis.upenn.edu

   Come on you guys. Have any of you really tried to use Lucid/GNUEmacs/ILisp/CLIM
   on a SPARCstation ELC or better (with at least 32 megabytes). I have been
   using such for almost a year now, writing tens of thousands of lines of code
   and find that the SPARCstation configuration is hands down better than
   anything Symbolics ever produced. The SPARCstation configuration is at least
   twice as fast as an XL1200 on the average and at least ten or twenty times
   faster than a 3640. For some things (like compiling source files using the
   development mode compiler and dynamic GC), the SPARCstation configuration is
   two orders of magnitude faster than an XL1200 (yes an XL1200). And it costs
   one fifth as much. And my ELC is a generation out of date. A SPARCstation-10,
   an RS6000, a MIPS R4000 or a DEC Alpha should make an XL1200 seem like a PDP-8.

   Ah, the development environment you say makes all the difference. But have you
   really tried Lucid/GNUEmacs/ILisp/CLIM? I'm not talking about Unix and C++.
   I'm talking about CommonLisp. Now don't tell me about hacking networking and
   window system internals since Unix provides all of that as a black box and
   most research and applications development work is not about doing operating
   system development. I find that at least 85% of all the wonderfull features
   of Genera that I used on a daily basis are available on my SPARCstation. And
   the ones that weren't there when I came along (like m-sh-C) I implemented
   myself. (If people are interested, I might be able to convince UPenn to allow
   me to release my implementation of m-sh-C for GNUEMacs/ILisp). The remainder
   are more than made up for by the blazing speed.

I have used exactly this config -- blech. Try to fix some of the
handshaking problems with a foreign editor to lisp. (I can do this but
the man hours you take trying to fix things up when you have code to
write.) Try to debug optimized code! I like speed, but at what price?
(One can do some amazing speed hacking on the Symbolics if you get into
the non Commonlisp level.) We really *still* need tags. The debugger
can drive you up a wall. Try to debug with several processes talking
to you at once -- all in the same window! (This is better now in some
lisp-emacs interfaces to be fair.)  CLIM streams don't like
interacting with the debugger still. FFI's can do a lot of copying of data
because the representation of data is different in C and tagged lisp
(Allegro or Lucid)/GNUEmacs/ILisp/CLIM is still pretty brittle.


   I was one of the first 3600 users. Back in 1982 I purchased two 3600s sight
   unseen, one of the first commercial organizations to do so. I didn't touch
   another brand of computer for ten years as I had the great fortune of having a
   Symbolics to work on everywhere I have been for the past ten years. So with
   great trepidation I faced the decision to switch to SPARCstations little over a
   year ago. It took me a few months to port several of my most current research
   projects totalling a few tens of thousands of line of code. Now in retrospect,
   if I was given the opportunity to switch back, even if you gave me a free
   XL1200 that was twice as fast, I wouldn't do it. Yes, every now and then I
   curse out my Unix machine for doing something in a brain damaged fashion and
   wish that the whole world did the right thing like Symbolics. But all I have
   to do to get over it is remember how I had to twiddle my thumbs when I typed
   <Select> L and my machine paged and GCed for the next few minutes. Really, try
   switching, I promise you that you'll never go back. I pity those poor souls
   still using 36xx machines.

I wish I still could and I don't need your pity, thanks. Speed
sometimes has a blinding effect. Remember that the memory a typical
36xx was ~4X2MW or 8Mbyte. They did run a *lot* better with more memory.

   Symbolics made one fundamental mistake. They ignored performance. They
   focussed their energy on making a whole operating system (ignoring all of the
   other things like Fortran/Pascal/C/Ada compilers, Joshua, Statice, Concordia,
   the S-products) when the real valued added of the their system was the Lisp
   development environment. Back in 1982 not only was the 3600 the best Lisp
   development environement, it also was the fastest Lisp execution vehicle as
   well as the one with the highest performance/price ratio. If they would have
   stuck to those three marketing goals: highest Lisp performance, highest
   performance/price ratio, and best LISP development environment, the areas of
   their unique strength, and did not attempt to reinvent the wheel of file
   systems, tape backup systems, network systems, window systems, etc. and
   instead leveraged off of other peoples work in those areas, they would probably
   still be in business today---and I would probably still proudly be using their
   products.

They still are in business.  I think the error was to hope to compete
with RISC and not consult with RISC folk to add tagged features. RISC is a
bit of self fulfilling prophecy - you get performance on what you
empirically use as target code.


IMHO, of course,
Albert Boulanger
aboulanger@bbn.com