[comp.lsi] Comp. Architecture survey

eugene@pioneer.arpa (Eugene N. Miya) (04/29/88)

[As one writer said:]
Most of the computers are really all about the same.

Depends on what you want:
 + most influential (whether or not they were "interesting" or contained
   any features not found all bundled together in one computer
 + unique, whether or not the unique features lived on in other
   computers
 + contained a feature that lived on in some important way
 + computers or processors?
[oh me:]
I asked for computers, got processors and software (summarized below).
Features or economics, I left for you to interpret (intentionally vague).
I really hate economic choices, but recognize them.

[As one writer said:] Here's my processor-centric list.
			(I may change my mind tomorrow.)
[I summarize interesting comments]

(18 votes) IBM 360/370 - first "architecture" and family of
binary-compatible processors; longest running big commercial success; horrible
software (read Mythical Man Month by Dr. Fred Brooks on the reasons--he was there).
Illustrates 'power of architecture', conformance to a scalable design
The great experiment in standardization. Nauseating though the architecture is,
the concept of a single architecture spanning a huge range of
price/performance and I/O subsystems was very important.
Well, probably third-place CISC if you consider the IBM 360 (which
is a lousy architecture by today's standards, but still breathing.
D. W. Anderson, F. J. Sparacio and R. M. Tomasulo; "The IBM System/360
Model 91: Machine Philosophy and Instruction Handling", IBM Journal of Research
and Development 11, 8-24; January 1967 (Is comp.arch gatewayed to the IBM world? ;-)
[Many books on this of course]

(17 Votes) PDP-11. They were the first computer for the (near-)masses.
Cheap and reliable, and DEC sold a LOT of them. for orthogonality/cleanliness,
elegance, addressing modes, stack support for OS and languages
also important because it was small and modular, and *everybody* made boards for it.
The addressing modes of this mini even influenced a major
early "inexpensive" minicomputer; early home of Unix;
It was very elegant, easy to understand, supported modern languages
very well, etc.  It is too bad that its 16 bit address space made it
obsolete only a few years after it was introduced.
[Many books on this of course]

(12 votes) CDC 6600 - polling instead of interruption (e.g. PPUs for I/O,
no flt. pt.  interrupts) and parallelism becomes useful (hardware scoreboarding)
the first widely used supercomputer
R. W. Allard, K. A. Wolk and R. A. Zemlin; "Some Effects of the 6600 Computer
on Language Structure"; CACM 7(2), 112-119; February 1964
[One good books on this]

(10 votes)
R. M. Russell, "The CRAY-1 Computer System", CACM 21(1), 63-72; January 1978
it works; fast scalar unit; prototypical "supercomputer".
Introduced vector processing [not really], and the idea of having huge
 numbers of registers in a machine (later copied by RISCs).
 Pioneered the concept of a horribly complicated machine
 to program.  Put compiler writers/researchers back into business.

(10 Votes) VAX : The cannonical 'ultimate' CISC, very symmetrical 
32-bit addresses, arbitrarily complex instructions with operand specifiers
interpreted at runtime (grind, grind, grind...). 9-11 cycles/instruction
* VAX 11/35 --> VAX 8800 And virtual memory!  Wow!
2nd-place for CISC == Success.  a major influence on RISC designers (e.g. disgust)
[Many books on this of course]

(7 votes) Apple ][ : Proved that micros could do fast graphics.  Proved that
people were willing to pay extra for compatability
[Many books on this of course]

(7 votes) Burroughs 1700/1800, 5000/5500, B5500, B6700, B7700
-- commerical stack machine, higher level language programming
a whopping 50 copies if I remember correctly(that number may actually
be as low as 30..) so you could argue that it wasn't a huge commercial success,
the first Machley/Echert (sp I Know!) award was given to Bob Barton.
(Algol) design base, one of the first virtual memories, multiprocessors,
software written entirely in a high-level language (1965), multiprogramming (in
1964!), a damm fine OS and user-friendly fault reporting system (source
language line number) -- variable microprogram machine
The fact that such a stack based CISC does not make sense today (as a *new*
architecture) is besides the point.
([A slight error:] Burroughs machines supporting Multics - influenced Unix)

(7 votes) Mac : Proved that you didn't have to buy an IBM PC.
commercially successful because of user interface;
set new interface standards and changed course of industry.
[Many books on this of course]

(6 Votes)  DG Nova -- arguably an early (1970) and popular RISC,
microprogrammed, right?) has had lots of imitators.
Soul of New Machine, T. Kidder - must read

(6 votes) DEC PDP 8 - successful mini machines, "laboratory"
a standard I/O bus running across the whole family, so that
new models could use the same peripherals as old ones.

(6 votes) IBM-PC : Proved people were willing to pay extra for support (or, for
the more cynical, for no reason whatsoever) Defacto standard [THIS IS THE
MACHINE PEOPLE WANT AS A STANDARD BENCHMARK MACHINE? HA!] scuttled by
orignal vendor(PS 2/OS 2) mostly of commercial rather than architectural interest
ubiquitous; revolutionized PC industry by creating a standard; now obsolete.)
they sold a lot of them.  Architecturally it's a mess.
(6 votes 1 weak) Honeywell 6000-series (not because of the machine,
but because of Multics).  (currently called the DPS-8 series),
66 "ICEbox", an instruction set expansion unit which made many COBOL
or PL/1 statements into single instructions.  This was folded into
the base architecture quite soon after its invention, and does make
it **quite** easy to write a certain class of code-generators for the
level 66 and 68 machines.
* ??? (later Honeywell) DDP-316, 516 -- the father of all 16-bit minis
(5 votes) IBM 701/704/709/709/7090/7094 - the standard scientific machine
7030 (STRETCH) - extravagance (bit addressable, etc.) a non-working supercomputer
architecture, prototypical CISC [36-bit] word machine
(5 votes) Ferranti/Manchester University Atlas.
They didn't call it virtual memory.  They called it a `single level store'.
influented a generation of UK engineers, & hence all modern ICL kit
[kit? THIS MAN IS A REAL BLOKE!]; pioneer (I think, but I'm a bit hazy hereabouts)
of "supervisors", index registers, associative registers, "extracodes", etc.
(5 (weak) Votes) The DG Nova had little impact on my career
(4 votes) Eniac
(4 Votes) Multiflow
(4 votes) Xerox Alto/Dorado : First workstation, graphics, mice, ..., shows what
you can do if you have enough hardware.
Alan Kay's original Dynabook paper - which presented the Smalltalk
Ed McCreight, Butler Lampson or any of the PARC pioneers
(2 votes - degree)Xerox Star - (was this the precursor [successor] to the alto)
(3 votes) 3. Manchester University MU5/ICL 2900 series. stack
and descriptor-based addressing.
(3 votes) Sun workstation family (Sun-2 (mc68010), Sun-3 (mc68020), Sun-4 (sparc),
Sun-386i (80386)) - leading workstation family; leading Unix platform;
birthplace of Network File System (NFS); first family built with 
language and operating system compatibility on diverse hardware architecture
(3 votes) "The connection machine" by W. Daniel Hillis is  a good 
 reference to someone having Physics mindset. There is a chapter in the
 book titled "Why Computer Science is No Good or New Computer Architectures
 and their relationship to Physics".
(3 (weak) votes and 1 anti-vote) Univac 1100
(2 votes) Amiga
(2 votes) DEC 10
(2 vote) "The ILLIAC IV Computer", IEEE Transactions on Computers C-17, 746-757;
G. H. Barnes, R. M. Brown, M. Kato, D. J. Kuck, D. L. Slotnick and R. A. Stokes;
August 1968
(2 vote) PDP-1 : Ran Space Wars, the first minicomputer. [Yeah I saw one at JPL.]
(1 vote) ALTAIR 8080 : The cannonical first microcomputer
(1 vote) Commodore-64
(1 vote) COM: (Caltech Object Machine) runtime binding microcode/macrocode opcodes
(1 vote) Cydrome
(1 vote) Ferranti/Manchester University Mark I.  `B-line modification'
(1 vote) Inmos transputer.
(1 vote) IBM 1620 - arithmetic by table lookup.
(1 vote) IBM 1401
(2 vote) GE 605, which evolved to the 615 and 635
(1 vote) IAS - the von Neumann concept implemented
(1 vote) Symbolics LISP Machine
(1 vote) MIT (data flow machine) (did not specify Dennis or Arvind machine)
(1 vote) Babbage's Analytical Engine - Historical
(1 vote) sds 940 [My DSN boss would have liked you]
(1 vote) Control Data Star 100 -- vector processor [Never been to LLNL eh?].
(1 vote) CDC 7600/Cyber 200/205/ETA 10 -- complex hardware
(1 anti-vote) NCR are easy to understand.  The designs were never published
and they were never used by people who are interested in computer design.
The NCR 8500 introduced in 1972 was the first commercial RISC computer.
It was used to emulate the instruction set of an older NCR computer.
But since it had no influence on anything besides other NCR computers,
it doesn't qualify to be on the list.
[P.S. my uncle Carl works at NCR on their computers some place...;-)]

I asked to get an idea of network biases and ideas about controlling survey
questions, and to confirm the level of mailing to postings (slips) which
I also did on the sci.space group (about same proportion).  The SDS just
slipped in time for instance.  The amount of wrong information is interesting
as well (I left some of it in).

I would have preferred a list similar to this fellow's:
Machine		Used	Read about	Reason
IBM-360/370,	x			First Main Frame
DEC-10/20,		x		First 'computer for people'
DEC-11/series	x			First 'computer for the masses'
Cray 1			x		Supercomputing
IBM 801			x		Risc
Apple II,	x			The peoples machine
Macintosh	x			The peoples interface

Machine which have impacted me personally: 1st and foremost the Xerox Alto
for the short time I used one at Caltech (roots to the DG Nova and
PDP/DEC-20/10/6), the CDC 6[7]x00 (1604 even), Cray-1, all for ideas.
Machines for numeric impact: IBM 360/370, IBM PC, Apple II, Mac.
For me not a great list. (I have used but found nothing special in:
Modcomps, Univacs, Varians, SELs, older H-Ps).

Rafael and I thought of a second list:
Apple III, PC Jr., TI ASC, Spectra 70, and some other machines, but some
of these architects read this list.

OTHERS note: things just as important:
Fortran(2), Algol(2), PL/I, Pascal(2), C(2), C++, Lisp(2), Prolog, Cobol,
Basic, Smalltalk(4), Unix(n votes I'm sure), VMS, OS/360, CP/M, (4 votes) Multics
A. Bensoussan, C. T. Clingen and R. C. Daley; "The Multics Virtual Memory:
Concepts and Design"; CACM 15(5), 308-318; May 1972

While not complete computers:
(1 vote) FPS-164
(1 vote) 32032
(4 vote) Zilog Z80 Z8000
(1 vote) 6800
(5 votes) 68000
(6 votes) Berkeley RISC David A. Patterson and Carlo H. Sequin;
		"A VLSI RISC"; Computer 15(9), 8-21; September 1982;
(5 votes) IBM 801
(3 votes) Stanford (RISC)
(2 votes) MIPS - move instruction scheduling to compiler
(4 votes Intel 4004/8008/8080/8086/80286/80386 -- silicon CISC
Backwards compatible with everything Intel has ever made
except the 432 and the 4004.  And it still runs!
(3 votes) Intel 432 -- The ultimate CISC == Horrible failure.
Effectively killed interest in capability systems.
(1 vote) Sun's SPARC.
I submit the Compaq 386/20.  Reasons:
	1.  Compatible with "standard" hardware/software,
	2.  Faster than IBM's new standard, even though similar pieces parts.
	    The question for your reader is: Why is the 386/20 faster than
	    PS/2-111?
		-  Memory access
		-  Display technology
		-  Disk sub-system access
	3.  It uses one of the most advanced CISC-style uprocs around.
	+. Haupage board pc clones - Haupage just make a 386 mother
		board that clone makers put in PC clones.  Has 64K
		cache, and 20 MHz model does ~4000 Dhrystones - and
		interesting price/performance point for $2500
* I consider the IBM PC to be only an incremental improvement over
existing machines.  It is mostly notable because (1) IBM knows how to
sell things (2) It came with a reasonable operating system (3) The
architecture and implementation was open enough that 3rd party vendors
could really go to it. (4) It was in the right place at the right time,
so the sold a couple.
(1 vote) Pixel Planes --  I admit personal bias on Pixel Planes.  I'm working
with the Pixel Planes team at UNC.  See Fuchs, et al. in Computer Graphics,
1985, no. 3 -- SIGGRAPH '85.)

Well thank these `experts:'
David Barto		Walter Bays		Tim Beres
Mike Butts		David Chenevert		David Collier-Brown
Bob Colwell		Lawrence Crowl		Mark Davis
Tim Donahue		Elliott S. Frank	Don Gillies
Bill O. Gallmeister	Marty Itzkowitz		Randell Jesup
Ralph Johnson		David Keppel		Bryan Lyles
Guy Middleton 		Henry McGilton		Robert Redford (uday@mips.COM)
David Schachter		Paul_L_Schauble		Steve Schlesinger
Mark Smotherman		Henry Spencer		Russell R. Tuck
Andy Walker		Steve Wilson	 Oh, and Rafael Saavadera
paul haahr

Wait a minute?  What about Seymour, Woz, Gene, Gordon Moore, Bill Wulf,
Gordon Bell, Steve Wallach, Baskett, Lampson, Wilkes, Steve Chen,
Alan Smith, and so on?
Aren't you glad that engineering isn't democratic at times?

Another gross generalization from

--eugene miya, NASA Ames Research Center, eugene@ames-aurora.ARPA
				soon to be aurora.arc.nasa.gov
at the Rock of Ages Home for Retired Hackers:
  "Mailers?! HA!", "If my mail does not reach you, please accept my apology."
  {uunet,hplabs,hao,ihnp4,decwrl,allegra,tektronix}!ames!aurora!eugene
  "Send mail, avoid follow-ups.  If enough, I'll summarize."

rminnich@udel.EDU (Ron Minnich) (05/02/88)

In article <8025@ames.arpa> eugene@pioneer.UUCP (Eugene N. Miya) writes:
>(7 votes) Burroughs 1700/1800, 5000/5500, B5500, B6700, B7700
>-- commerical stack machine, higher level language programming
>a whopping 50 copies if I remember correctly(that number may actually
  Just one comment here, the 17/18 series was quite a different
machine from the 5/6/7 series. In fact the 17/18 was quite a 
departure for Burroughs, as it was a 'user-level' microprogrammable
machine, and supported a Cobol set, a Fortran set, and others.
   When i worked at Burroughs i was told that the machine was 
deliberately crippled so that it would not put the 4x machines 
out of business (sound familiar?). The clock was cut and a couple
other things done so that 4xxx machines would not lose sales. 
The machine was orphaned almost from the start. Kind of a shame.
   At some point there was supposed to be an effort to build
a C set whether RISC-y or not i don't know. This was before RISC
so i doubt it. The project died when Burroughs killed a lot of
projects in the early 1980s.
ron
P.S. The 4xxx machines were BCD with cobol support instructions.
-- 
ron (rminnich@udel.edu)