jeh@ritcv.UUCP (07/13/84)
I just finished watching a show from Ontario Province called "Bits and Bytes".
It was shown on my local PBS station. In this episode, a man plays the role
of asking questions, and an apparent computer trends expert answers them. The
topic was computers in education:
Question Man: Programming? Me? Well, I did try a little BASIC once, and
that didn't seem to hard, and then I saw LOGO was even easier!
Is there anything coming up that is going to make our home
computers even easier to program?
Answer Woman: As a matter of fact there is. There is a new language that
will soon be available to micros. It's called...
Me, thinking: [Oh, God, she isn't going to say Smalltalk, is she?]
Answer Woman: ... "Smalltalk".
Me, thinking: [Oh, GOD! She said it!]
:-)
Anyway, the cameras then switched over to Adele Goldberg (by the way,
Goldberg.PA, if you're reading, congratulations on your new job as
president of the ACM!) telling us how great Browsers are because they
let you find interesting things while you are looking for something else.
We also saw some Xerox machine (probably a Dorado -- your typical home
computer micro) running Smalltalk-80.
To say the least, I was shocked. Does anyone out there want to defend
that prediction of Smalltalk-in-the-home-soon?
Jim Heliotis
{allegra,seismo}!rochester!ritcv!jeh
rocksvax!ritcv!jeh
ritcv!jeh@Rochester
P.S. Are a lot of UT people into Monty Python - type cartoon feet?rej@cornell.UUCP (Ralph Johnson) (07/17/84)
If you look at the recent Smalltalk papers (such as the ones in the
latest POPL or the paper on Berkeley Smalltalk in the latest SIGSOFT
(the last may be wrong)) it is obvious that current opinion is that
several megabytes of main memory is needed for Smalltalk. However,
"Bits of History ..." refers to versions of Smalltalk for Z80 and 8086
class machines. The Alto was not a very powerful computer - the Mac
should be able to run rings around it. If these machines could run
Smalltalk, why couldn't a current micro?
One thing to note is that most Smalltalk machines are inefficient. The
ports to 68000 machines reported in "Bits ..." are incredibly slow, one
needed five second to echo a character! However, implementations are
getting better. Peter Deutsch describes a resonably efficient version
in the latest POPL proceedings. In addition, many micro languages are
inefficient. Smalltalk might make it as a cute toy, even if it were
too slow for serious work.
The biggest impediment to wide use of Smalltalk in the past has been
Xerox, since they did not let any information about it out. Now that
they have let books be published and have even released the Smalltalk-80
virtual image (for a reasonably hefty fee) there is more chance that
Smalltalk will make it.
Ralph Johnson {decvax!ihnp4!vax135! ...}!cornell!rejbrucec@orca.UUCP (Windows on the World at ECS) (07/17/84)
-------- There was a varient of Smalltalk-72 built by a company in Texas unconnected with Xerox, called Rosetta Smalltalk. It ran on the 8085, and was later ported to the (gasp!) iAPX-432. I used both for a while, and was convinced that: 1) Rosetta Smalltalk was not as powerful as Smalltalk-76 (and in retrospect, it was nothing like as useful as Smalltalk-80). 2) Smalltalk is a pig on a small machine. The 432 implementation had 256K to play with, but a fatal hardware design problem with the processor board it ran on made it intolerably slow. Good hardware, and the ability to run multi-CPU might have made the 432 Smalltalk (Intel sold it as OPL, for Object Programming Language) a reasonable demo for the chip and the language. Recent experience with honest-to-Goldberg Xerox Smalltalk-80 on a Tektronix Magnolia (there's a picture of it in the orange book), which is a 68000-based system, and is usually described as being about half as fast as a Dorado, have convinced me that Smalltalk is a viable programming language ON A REASONABLE PROCESSOR. The problem is that there aren't many micros which have the combination of features which make them really efficient for implementing the Smalltalk virtual machine. The 68000 works, but it needs a lot of help, in terms of support hardware, and that costs money. The Mac was built for large-scale, low-cost manufacturing, not high performance (the real killer is the way it handles the screen; it's quite elegant, and very cost-effective, but it would probably cut Smalltalk's throat on BitBlT). Smalltalk really needs large memories (> 1 Meg) and a hard disk, which adds quite a bit of money to any system. There's an apocryphal story I heard a couple of years ago, that the DEC implementation of Smalltalk-80 on the VAX used an entire 11-780 (single-user), and ran at 20 Kbytecodes per second, until it hit a BitBlT, at which point it went down to below 1. By contrast, Apple says that its Lisa Smalltalk ran at > 5 Kbytecodes/sec, and the Dorado is supposed to be capable of >30K. I think that Smalltalk is a lovely system, as long as it has reasonable response. Right now that means at least $15K, but in two years or so it will be much less. Bruce Cohen UUCP: ...!tektronix!orca!brucec CSNET: orca!brucec@tektronix ARPA: orca!brucec.tektronix@rand-relay USMail: M/S 61-183 Tektronix, Inc. P.O. Box 1000 Wilsonville, OR 97070
mdrutenberg@watmath.UUCP (Mike Rutenberg) (08/11/84)
Date: Wednesday, 18 July 1984, 1:51:06 pm
From: Deutsch.pa@Xerox.Arpa
Subject: Re: "Smalltalk Coming to Micros!" ?... HA!
Just to correct some errors of fact: according to the published
benchmarks, using the sum of the times for the macro-benchmarks
as the overall performance figure, Smalltalk-80 runs at
approximately 15-20K bytecodes/second on the Lisa, 30-40K on the
Dolphin, about the same speed on the Sun Microsystems
workstation (68010) under the U.C. Berkeley implementation (BS),
70K on the Magnolia (this is a private communication, better than
the published number), 100K on the experimental 68010
implementation at PARC (as reported at 1984 POPL), and
250K-300K on the Dorado. The Berkeley and PARC 68010
implementations use 32-bit pointers; all the rest use 16-bit
pointers. As of this writing, only BS has a working, usable
virtual memory system (it runs under 4.2bsd): I don't know
whether its published performance figure is CPU time or real time.
[[Note: Net.lang.st80 is now somewhat (that's me!) linked with
a Smalltalk-80 interest list on the Arpanet. Posting to Usenet
will make it and the other way may make it (snicker!). If
you've seen this before, I'm sorry (I don't think it made it
out of uw-june before it went down for 4.2).
A reply path to Peter is:
Usenet: {allegra|decvax}!ucbvax!Deutsch@Xerox
Arpanet/CSNet: Deutsch@Xerox.Arpa
If anyone has troubles, I'll try to help (mdr@uw-june)
Mike Rutenberg
Xerox Parc
(415)494-4379]]