\
Gordon
Bell, vice president, Engineering
by Richard Seltzer, from DECWORLD the company
newspaper, July 1983
Looking
ahead -- The excitement has just begun
"I used to consider myself to be a computer
architect, working on individual systems. Now I believe I'm more
like a city planner. I worry about how the nodes are connected and
how information flows between them and how one can build from a
collected set of nodes. I think about local area networks and how
they are connected to form wide area networks, how technology such
as cable TV is used, and how to get the necessary bandwidth and
standards, since with distributed computing the need for
communication grows exponentially with the number of computers.
While we need to build larger, more reliable computer systems, it
is becoming possible to it by interconnecting the emerging
super-micros rather than continuing to make larger individual
machines.
"Economy of scale is disappearing, and it's
much cheaper to engineer and manufacture smaller computers to get
the best cost-performance. But we need to understand how to use
them together in all kinds of structures. For example, the beauty
of Ethernet is that anyone can link the computers together, and
segments can easily be connected to form a whole at a single site.
"Today we're building more complex systems, but
they're built in an orderly way. Today's basic component is a
processor -- a whole computer. We have to cover a lot more levels
of integration now than we ever did -- from semiconductors to
applications.
"As the company's second computer engineer, I
didn't have a notion of architecture. When we defined something,
it was in one person's head. One engineer designed the circuits,
and somebody else designed the logic. There were a few diagnostics
and systems programs. The user had to link together various
subroutines to perform a unique system -- rather than having an
operating system.
"Now today, with integrated circuits, we are at
the point where there is more creativity for everybody including
the circuit engineer, where the problem is laying out circuits on
a chip. Architecture is refined so that a machine has a meaning
over seral implementations as we've done in the VAX and PDP-11
families.
"I recently visited a user in Australia who
built a totally homogeneous VAX computing environment: with large
VAXs in Sydney running central computation centers and distributed
730s in various regional offices. They also had networked control
computers doing real-time applications. That whole system ran as a
single computing environment, along the lines I had hoped for with
the VAX and a homogeneous architecture.
"What we're doing with networks and
multiprocessors -- making it possible to run a single problem on
many machines -- creates a new domain of architecture, which is
more exciting than anything we've had in the past.
"Meanwhile, the exciting applications in areas
such as Artificial Intelligence on parallel machines are still
ahead of us, and probably won't really get going until the early
1990s, even though there are breakthroughs every day in specific
knowledge- and expert-based systems.
"It continues to be clearer and clearer that we
are only at the beginning of the computing era."
Lessons
from history
Gordon has an ever-expanding set of notes,
heuristics and lectures that will eventually become a book on the
evolution of computing. He points out problems that could be
avoided if engineers would pay closer attention to the lessons of
history. "Often a product starts out simple, then becomes elegant
(a design is 'elegant' when one part is used for two functions).
Then it gets extended to be more general, and finally ends up so
general that it's tricky. If it's too simple, it may be naive and
useless.
"We also se that ideas tend to go in cycles.
For instance, in a process known as the 'wheel of reincarnation,'
you first add some simple hardware to a product to do a particular
function such as displaying information. Then you put in more
special hardware to do the function and to offload the original
information processor. Gradually, you add more and more hardware
so the function can be done better and without any help from the
main processor. Ultimately, you end up with a general purpose
processor which does the display function and you've totally
unloaded the first processor. So you really end up with two
separate computers, neither of which is doing very much, except
communicating with the other one. This has happened in graphics,
communications, computation and database processors.
"History also helps give us a sense of
perspective -- of the importance and potential of the products and
ideas we've grown accustomed to working with. Since we're so close to
specific products, one of the hardest things to know is when we
should stop evolving one and take a new approach. I think history
provides good lessons on determining when a product is
over-the-hill in terms of extensibility and hence lacks
competitiveness. I
strongly believe in product euthanasia."
Engineering
around the world
In the last few years, Gordon has sponsored
engineering groups on the West Coast, Colorado, the U.K. and
Japan.
"The West Coast and Japan have significant
groups of talented engineers. We want to be able to take advantage
of the technology that's in the Silicon Valley and in Palo Alto
around Stanford. Also, Japan has superb semiconductor, display,
magnetic recording, printing and consumer electronics technology.
The Seattle location has a group of engineers who wanted to
relocate from New England. They work together in a small,
effective, tight-knit group. Colorado Springs started out the same
way. In terms of productivity per person, they're both
outstanding.
"In general, I have little faith in large
design groups, and virtually no faith in large committees to solve
problems," he explains. "Large groups tend to spend a lot of their
time talking and politicking with each other rather than using
interface specifications and the written word. In contract, one of
the most complex interfaces we've developed is the mass storage
control protocol that links mass storage devices with VAX systems.
That was worked out very precisely between engineers in Colo9rado
and VAX/VMS engineers in New England, totally through written
specifications, and when it was plugged together it worked.
"The only engineering DEC has in places like
Australia and France is done by Computer Special Systems (CSS). These small groups are
tightly linked to customer needs.
Central Engineering has not yet been able to take advantage
of this worldwide engineering talent as CSS has done. I think
that's something that I'd like to see us do int he next five
years.
"For example, several Australian CSS engineers
built a statistical communications multiplexer. We're shipping about
1000 of those per year. That's enough business for a
respectable-sized small company, and the project cost less than it
costs to write a business plan. That's another reason why we need
smaller engineering groups."
using computers for design
The most important recent change in Engineering
has been the new emphasis on the use of computers to help design
better computers. "The
use of computers in computer design really didn't become mandatory
until the mid-70s," notes Gordon.
"But now we can't do without them for VLSI and large
systems design. We use computers to lay out circuits to simulate
the operation of circuits, and, finally, to test the actual
devices.
"Our increasing use of computers in engineering
requires changes in training and changes in behavior. We call this
approach the "Quality Design Methodology." It is being used in
VLSI, large system development and some of the mid-range systems.
Other areas are in the third generation or dark ages and must be
retrained along the new lines.
"This discipline is one of the indirect
benefits of 'silicon Mountain' (our Hudson, Mass., facility for
designing and manufacturing large scale integration -- LSI --
semiconductor circuitry). They are working on vital technology
which we're going to need for designing products. Within ten years, some
systems we're going to build will be on a single chip, and we must
master VLSI. This demanding work requires clear understanding of
fundamentals, including materials. It has taught us a great deal
about managing complexity (i.e., systems with lots of
interconnected parts), and that's something that every engineering
group must eventually understand.
"I expect every product breadboard to be so
well designed and so well constructed that it will work the first
time it is powered up. This
means: having a clear organization; understanding the behavior of
all supporting processes; specifying the design in more and more
detail so that at each stage it has no errors; inspecting the
design rigorously; simulating the design and verifying its timing
using computer-aided design (CAD) tools; building it and getting
ack a breadboard that is virtually perfect because Manufacturing
uses a similar methodology; and then finally applying the power
and observing that the actual device behaves exactly as the
simulated version."
Looking
back -- MIT, Australia and the start of a career in engineering
Gordon's first exposure to computers was as an
undergraduate at MIT, where he learned programming on the
Whirlwind, a computer that Ken Olsen and Dick Best had helped
design.
After graduation, he received a Fulbright
scholarship and went to the University of New South Wales in
Australia. It was a new university with a new British computer
called the DEUCE, that Turing helped design. During his year there,
he wrote software and helped organize and tech their first
graduate course on computer design.
"When I came back to the U.S. in late 1958,"
explains Gordon, "I had a choice of going to work for Philco on
computer design or returning to MIT. I decided on MIT where I
could do work that had a very high computing content as opposed to
a circuits and logic design orientation.
:My thesis advise, Ken Stevens, who ran (and
still runs) the Acoustics and Speech Research Lab at MIT, asked me
to work on their new computer, the TX-0 given by Lincoln
Laboratory. I interfaced speech input equipment to the TX-0 and
wrote analysis and recognition programs (one of which, Analysis by
Synthesis, is still a mainstay of recognition programs). We needed
secondary memory; so I got a tape deck, interfaced it to the TX-0,
and in the process discovered DEC, a new little company that was
starting to sell system modules.
"these modules were compatible with the TX-0,
because that computer was really the breadboard for the Lincoln
Laboratory transistor circuits that Ken Olsen and Dick Best
developed, and those circuits were really the breadboard for DEC's
first modules. I went to Maynard to buy modules and met Ken, Dick,
Harlan Anderson and Ben Gurley.
"Speech understanding and recognition is a
classic hundred-year-old problem. It's deceptively simple because
humans do it, and humans can also recognize speech represented in
two-dimensional images called voice prints. I thought I could
write a program to recognize speech. It took me a year to
understand that this problem would take 20 years to solve, and I
don't like to work on one problem that long. It was really
computers that fascinated me.
"Besides, I didn't want to be a researcher; I
wanted to be an engineer and build things. DEC looked like exactly
the kind of place where I wanted to work. So in 1960 I gave up my
doctoral program and came to DEC.
The PDP-6 development team in 1964. Left to
right, standing: Peter Samson, programmer; Leo Gussell,
diagnostics programmer; Gordon Bell, system architect; Alan Kotok,
assistant architect and logic designer; Russ Doane, circuit
engineer; Bill Kellicker, technician; Bob Reed, technician; George
Vogelsang, draftsman. Seated: Lydia McKalip, secretary; Bill
Colburn, technician; Ken Senior, Field Service technician; Ken
Fitzgerald, mechanical engineer; Norman Hurst, technical writer;
Harris Hyman, programmer.
Early
days at DEC
"I was the second computer engineer in the
company. Ben Gurley, who headed Computer Engineering, hired me,
and Dick Best, who is still our Chief Engineer, was in charge of
designing modules. I was hired to do programming, architecture and
logic design.
"Tom Hastings was probably DEC's first software
engineer. I hired him as a summer student in '61 to work on the
PDP-1 program library. (Tom now works in Terminals in the Mill and
has made important contributions int he architecture and standards
of terminals. He also played a significant role int he PDP-10 and
the VAX.)
"DEC got started in the computer business at a
time when the industry was shifting form vacuum tube technology
(what I call the first generation of computers0 to
transistor-based technology (second generation). The PDP-1,
introduced in 1960, was one of the first commercial applications
of that technology.
"A total of 50 PDP-1s were built. We stopped
making that product around 1965.
One of the luckiest things that happened was that half of
the machines were sold to ITT for message switching. It could handle up to
256 telegraph lines and switch them to other telegraph lines.
Working on that project, I developed the device which we now call
the UART (Universal Asynchronous Receiver Transmitter), a serial
line encoder-decoder.
"We were very lucky in getting the ITT order
because that made the PDP-1 a standard product. If that hadn't
happened, I don't' think we would have survived int he computer
business.
"Every other order for the PDP-1 used a wide
variety of custom-designed and custom-built options. Those were
the days when you first offered something for sale; and then if
someone bought one, you designed
it. That's a crude way to do market surveys.
"One of our earliest customers, Lawrence
Livermore Radiation Lab, sued their PDP-1 as a converter between
the IBM STRETCH and UNIVAC LARC. They ordered a system with
virtually every option we could think of (but had not yet
designed) including UNIVAC and IBM tapes, a UNIVAC card reader, an
IBM card reader and an IBM card punch. They even ordered a five
inch scope with 4096 x 4096 resolution that's still unrivaled
today in its precision. We were lucky we didn't have more orders
for such exotic equipment.
"DEC's goal from the very outset was to build a
whole scale of computers -- from small to large. But our approach
was rather haphazard -- a far cry from today's methods.
"The PDP-2 was a mythical machine number
reserved in case we wanted a 24-bit computer. It was never defined
on paper. The PDP-3,
a 36-bit computer, was defined; and one of our customers built one
using DEC modules. We almost got an order for a PDP-3 from the Air
Force Cambridge Research Lab, but Harlan Anderson, DEC's first
vice president, and I persuaded
them to take two PDP-1s, in other words two 18-bit machines
instead of one 36-bit machine.
"The PDP-4 started us in a new line of business
for real-time control. It was designed for process control and
ease of interfacing for a couple of Foxboro's customers. Because we didn't
understand the cost or value of software, it turned out to be a
business mistake. But, fortunately, it led to the PDP-5, which in
turn led to the PDP-6 and the start of the minicomputer industry.
:One of the first few customers that bought the
PDP-4 was the Canadian Atomic Energy commission at Chalk River. They wanted to control a
nuclear reactor and needed a machine to do front end work for
reactor monitoring. We visited them to consider the monitoring
requirements. We were thinking of it as a special system. But in
looking at it, we came upon the idea of the PDP-5 -- a computer
designed to do process monitoring and recording and to work with
the PDP-4.
"I think everybody who builds specialized
digital systems discovers that the best digital controller is a
computer. So no matter what we start out to build, we always end
up with a computer. I
continue to see engineers, companies and industries stumble onto
this -- most recently in the form of microcomputers.
"Some people call the PDP-5 the first
minicomputer, but I don't think of it that way, even though it was
built to be embedded in a larger system.
"I believe the PDP-8 is the first real mini. It
was half the cost of the nearest competitor. It was also a lot
easier to interface, much faster and far smaller than the PDP-5.
It took less than half a 19" rack, so you could embed it in any
system. In other words, the computer was clearly a component.
"I like to think that minicomputer stands for
'minimal' computer -- that is the smallest computer you can make
at a given time for the lowest price: something you use to
substitute for other digital systems. I think that micros are
really mini(mal) computers.
"The whole notion of personal computing is
something that DEC pioneered. Personal computing to me means
having a screen, very good interaction and a filing system so that
the machine takes care of the bookkeeping of programs and data. It was a tradition that
we all learned about at MIT. The TX-0 had interactive use that
included a cathode ray tube (CRT) and typewriter. It turned out
that other computers had typewriters, but they didn't have a CRT
and usually didn't have a good file system. That's why I think
that the first real personal computer was the LINC, designed by
Wes Clark and Charlie Molnar at Lincoln Laboratory. It's no in the Computer
Museum.
"Designed with DEC modules, the LINC had its
own file system called "LINCTAPE." Ultimately Digital manufactured
it. "When the designer of the file system, Tom Stockebrand, came
to Digital from Lincoln Laboratory, LINCTAPE evolved into DECTAPE,
and people were able to have personal filing systems, write and
program directly on the CRT, compile and then execute them without
having to go through a central data processing facility. Also it
could be moved from lab to lab. No other system offered this
capability until we started timesharing systems.
"Minis and micros are one kind of personal
computer that we pioneered. Timesharing is another. By timeslicing a
computer among many users,. we provided each user with what
appeared to be his or her own large computer. This concept was
implemented in three different places at the same time. Bolt
Beranek and Newman (BBN), a management consulting firm in
Cambridge, Mass., one of the earliest customers for the PDP=1,
ordered two typewriters with their first system and did some of
the first work on timesharing. Meanwhile, MIT was using several
typewriters on a single computer, and Systems Development
Corporation built a timesharing system on a large military
computer.
"It was from these ideas that we created the
PDP-6, the forerunner of the DECsystem-10 which was built from
scratch to be timeshared so that everybody could have their own
large computer for personal computing.
Gordon Bell (left) and Alan Kotok (right)
working on the PDP-6, circa 1964.
Time off
for research, then back to DEC
"In 1966 I went to teach, learn, write and do
research at Carnegie-Mellon University, still staying in close
touch with DEC as a consultant and a customer. While there, I consulted
on development of what became the PDP-11 family of computers.
"The main reason that I went to Carnegie was to
get time to think about computers. In retrospect, the leave was
perfectly timed. I left a the beginning of the third generation of
computing, when integrated circuits were coming in and every
garage and store front was being converted into a minicomputer
company building. At that time, the main issue at DEC was
switching from discrete circuits designed and built in-house to
circuits that semiconductor companies built. Aside from their
smaller size, there wasn't much difference in working with these
new circuits. The transition took about six years -- the time I
was at Carnegie.
"In 1972, I was going to take a sabbatical and
write another book, the 'ultimate' one on designing digital
computers. But Ken
Olsen asked me to come back and be vice president of Engineering.
"I returned to DEC for several reasons. Large
scale integrated circuits had just become available, and the
notion of a processor on a chip was beginning to appear. That was
the start of the fourth generation of computers. I had done a lot
of work at Carnegie based on the belief that processors and logic
were going to become very cheap. All the multi-processor and
multi-computer work at Carnegie was predicated on this. I came back to get DEC
interested in LSI and ready to start projects based on it.
:I also wanted to work on what amounted to VAX.
At that time it was clear the PDP-11 wasn't big enough. We had to
do something else, and I was interested in that next step.
:?but probably the most important reason was
that I had spent a lot of time trying to understand computing as a
computer scientist, and I really wanted to be directly involved in
building machines as an engineer.
"In 1972 there was no such ting as Central
Engineering. There's always been a central core of engineering
people. But back then, they were distributed in various groups
called 'product lines,' that had both marketing and engineering
functions and operated in a profit and lo9ss framework.
"In 1974, we brought most of engineering
together as a single group. From
the beginning, we said that Central Engineering would concentrate
on basic rather than market-specific products. The product lines
retained small engineering groups dedicated to tailoring these
basic products to meet the needs of their particular markets.
:We soon got projects going in the LSI area,
especially the LSI-11 with Western Digital. VAX-11 came later.
That's the name I used beginning in April 1, 1975 in the task
force to extend the addressing of the PD-11. It originally stood
for 'Virtual Address Extension-11' a reminder that we were
extending the 11, not starting over. That's a case where the name
stuck throughout the project's life and has become one of our
major trademarks. The press discovered it prior to announcement.
So, when the time came to settle on a public name for it, we
stayed with it, because we had already had more press exposure
than if we had planned a campaign.
"Today, Engineering is more aligned with
Manufacturing and has the marketing functions for the basic
products it builds. Every product has a business plan which it is
measured against, and the engineer who plans or proposes a product
is responsible for doing it. The express 'he who proposes does'
reflects this notion that planners and proposers aren't distinct
from implementors. Today Engineering has more responsibility than
it has even had because it is solely the project team that I hold
responsible for a product."
seltzer@seltzerbooks.com
privacy
statement