Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!watmath!clyde!cbatt!cbosgd!ihnp4!ima!johnl
From: johnl@ima.UUCP (John R. Levine)
Newsgroups: net.arch
Subject: Re: top down vs. bottom up design [was strange sex]
Message-ID: <166@ima.UUCP>
Date: Sat, 19-Jul-86 15:08:38 EDT
Article-I.D.: ima.166
Posted: Sat Jul 19 15:08:38 1986
Date-Received: Sun, 20-Jul-86 06:19:26 EDT
References: <2900019@ztivax.UUCP> <802@tekig4.UUCP>
Reply-To: johnl@ima.UUCP (John R. Levine)
Organization: Javelin Software Corporation
Lines: 48
Summary: the issue is can your design be reimplemented

In article <802@tekig4.UUCP> jerryn@tekig4.UUCP (Jerry Nelson) writes:
>In article <2900019@ztivax.UUCP> david@ztivax.UUCP writes:
>>I know of one system which was completely developed [software first.] Some
>>software people wrote "the perfect language" and the "perfect OS
>>concepts" and then some smart HW folks developed the hardware to
>>support it.... It has got to be the world's most un-portable system.
>>David Smyth
>Hold it!  Are you telling me that there really is such a thing as portability?

This isn't portability, this is reimplementability.  The IBM 360 instruction
set has been implemented at least 25 times, with a performance range of at
least 1000 between the slowest (360/25, ca. 1965) and the fastest (3090,
ca. 1985 or maybe the Fujitsu engine.)  The PDP-11 instruction set has been
implemented about a dozen times, more if you count PDP-11 mode on a VAX.
The 360 and PDP-11 were designed more or less software first, although in both
cases changes in hardware technology and software technology have made the
architectures look a little tired by now.  In both cases you can take object
code from the early models and run it on more recent models and it still works
(by and large -- I'm thinking more about user programs than operating systems
here.)

But this begs the point.  People aren't implementing the PDP-11 much any more,
and there are good reasons for that.  When they designed the '11 about 15 years
ago, it seemed that 16 bits of address were a lot for a computer of that size.
What do you know, memories grew faster than they expected.  The 360 has the
same problem, although they can be excused a little by noting that they started
a lot earlier, and they never dreamed that their architecture would still be the
dominant one over 20 years later.  (They can also be cursed because the 360's
addressing was shrunk from 32 bits to 24 by what were obviously warts, even
at the time.)

It's perfectly wonderful to design your computer so that it is well matched to
the software that is going to be run on it, but your computer will go nowhere
if its implementation is static.  All of the successful architectures have had
multiple implementations, going back to the IBM 704 in the mid 1950's, and
that's not going to change.  I've seen precious little design that is set up
to work smoothly in the face of reimplementation, except maybe for Intel
encouraging us to pretent that the 8086 had segments because subsequent
reimplementions really would have them.

I'd be interested in hearing from the RISC crowd -- how much to they tune
their designs to the precise technology available today, and how much do
they expect to carry over into their next generation?
-- 
John R. Levine, Javelin Software Corp., Cambridge MA +1 617 494 1400
{ ihnp4 | decvax | cbosgd | harvard | yale }!ima!johnl, Levine@YALE.EDU
The opinions expressed herein are solely those of a 12-year-old hacker
who has broken into my account and not those of any person or organization.