PDP-11: All Out to Win (Tenth Anniversary of a Computer Family)

by Richard Seltzer, from DECWORLD, the Company Newspaper, May 1980

 The year was 1969, the 12-bit PDP-8 was the minicomputer, and DEC was the PDP-8 company.

 Several firms had already introduced 16-bit machines, and customers were anxiously awaiting DEC's answer to the competition. Internally, there was an understandable resistance to a product that might displace PDP-8 business.

 The prospect of a 16-bit computer had been discussed for several years. In fact, two projects had been started. The first was abandoned because it failed to meet cost goals. The second, code-named DCM (Desk Calculating Machine), had run into design problems.

 Two central processors -- an 8-bit and a 16-bit -- were partially designed. An 8-bit memory was ready for release. Software was under way. Packaging had been designed. A common data bus would tie together central processor, core memory and peripherals as system units that could easily be replaced or added to. But the processor design was not crystallizing. Benchmarks were run and instructions added; one change led to another and another, with no end in sight.

 Finally, in March 1969, the decision was made to switch to an entirely new 16-bit architecture that simplified programming and took full advantage of the common data bus which became the UNIBUS.  Work then proceeded at a frenzied pace to get a machine ready for market. DEC was coming from behind. The pressure was on.

 In April 1970, a little over a year after adoption of the new design, DEC delivered its first PDP-11 -- not just another 16-biot computer, but the start of a family would grow at a phenomenal pace. Now in its fourth generation, the PDP-11 has set standards of performance and reliability for the minicomputer industry and has transformed the company. Due in large part to the success of the PDP-11, DEC has grown 20-fold over the last 10 years.

 Design born in March

 Although small in comparison with today, DEC, in 1969, was the world's leading minicomputer manufacturer. It had a well-established reputation, nearly $100 million in revenues and about 4000 employees worldwide. That spring only a couple dozen people in a corner of the Mill were working on the PDP-11, just one project among many that were then in development.

 About Christmas 1968, Roger Cady had taken over the DCM project from John Cohen and renamed it "PDP-11". Hal McFarland and Bev Young, key engineers, were already in place. Cady brought in Jim O'Loughlin to help develop the central processor, Paul Janson to work on the UNIBUS and peripherals and Ken Hedberg to get things done. He also invited Bruce Delagi, Rick Merrill and others to consult on the architecture.

 "McFarland had proposed a completely different architecture for the instruction set processor.  Time was important; it seemed too late to make such a wholesale change. On the other hand, the architecture we had was not working out." Cady recalls.

 So one snowy Sunday morning, March 9, 1969, Cady, O'Loughlin, McFarland and Merrill flew to Pittsburgh to meet with Gordon Bell and Bill Wulf, professors at Carnegie-Mellon. Bell had been consulting on the DCM project, and McFarland was a former student of his. The consensus of that Sunday meeting was to go with the new architecture proposed by McFarland, a decision later ratified by the Operations Committee in Maynard.

 "We wanted an architecture that would last. So in the course of about a week, we decided to discard about a year's worth of work," O'Loughlin remembers.

 Dave Nevala and Chuck Blasi were making progress on the packaging. Programmers under Hank Spencer and John Hittlell were busy writing system and diagnostic software. Bob Hamel and Pete Durant were working on memory. With the new design, most of them had to turn around and start all over.

 O'Loughlin worked all spring on logic design of the central processor while others tuned up the instructions and made final changes in the UNIBUS.

 Bell had advocated the use of a single bus architecture as opposed to the separated bus structures that had been used in all previous computers.  This concept, embodied in the UNIBUS, was an important part of the compute. The first single data bus to enable minicomputer devices to send, receive or exchange data without processor intervention, the UNIBS provided an unexpected bonus for the design team. It gave an interface specification for each piece of the system. This meant that memory, peripheral and central processor development could all proceed independently and in parallel. "That enabled us to go faster, to deliver system units modularly and expand as we went along," explains Cady.

 By April3, Cady had sketched out plans for a family of eight machines, cover a wide performance range.

 Suddenly, the project shifted gears. "Sooner or later you have to stop designing or you never get products," explains Cady. "You have to say, 'No more discussion. Enough inventing. Now we product.'"

 Starting from a brand new design in march, the team produced experimental wire wrapped models of the PDP-11/20, the first member of the family, that summer; printed circuit prototypes that fall and machines ready for delivery to customers the following April

 "We did in 13 months what now takes 18 months in two years," notes O'Loughlin.

 There was a great deal more informally 10 years ago, said Bob Puffer, then in charge of manufacturing engineering. There was very little stratification of effort. managers were designing, technicians were designing, designers were managing and everybody was expediting. Jobs wee parceled out to people who had time. Relatively little was systematized or written down. Much happened verbally in face-to-face conversations -- people walked around with engineering notebooks under their arms, had impromptu meetings in hallways, jotted own notes and went off to make things happen.

 Engineers wrote their own component specifications, Cady remembers. Drafting, printed circuit layout, the board shop and other support operations were all near at hand int he Mill so it was relatively easy to get things done and make changes.

 On the other hand, proximity to manufacturing meant noise. the DCM group had been up in Mill 5-5, trying to design to the tune of wire-wrapping guns on the PDP-9 production line. Cady moved the group down to 5-2 -- somewhat quieter, even though it was right below the loading dock.

 Tom Karpowski became Field Service's product support manager for the PDP-11, and technicians from Field Service and Quality Control joined the engineering team, not just to look after the interests of their groups but to help design, build and debug the breadboards and prototypes. For the debugging process, the whole team went on a three shift basis; but even at that, 60- and 70-hour weeks were common.

 The Operations Committee kept close watch on every move. Design reviews were held in Ken Olsen's office. The company was committed to producing this computer and getting it to market rapidly.

 Despite time pressure the PDP-11 engineering group championed changes int he company's design and manufacturing practices to make the product economical to manufacture and service.

 In particular, to implement the modular concept (system units tied together by the UNIBUS), the designers opted for what were then considered large densely packed printed circuit boards.

 The company had previously built computers using backplane wiring and standard flip-chip logic cards. The PDP-11 group used that technique to build the first working models. Then, module by module, as the design checked out, bulky assemblies of those standard cards were replaced with dedicated printed circuit boards etched specifically for the PDP-11.  The computer kept getting smaller and smaller. By the time it was done, the entire PDP-11/20 fit into a box 10-1/2" high -- about 1/12 the size of the original mock-up.

 Memory cores were put down flat on large boards, and wire were run through in one pass rather than stacked in layers then wired together. Advances such as this helped significantly to reduce hardware, assembly and service costs.

 The circuit boards presented another challenge to manufacturing, which was set up to produce flip-chip modules a quarter their size with wider circuit lines. But changes that seemed drastic at the time soon after became the accepted norm.

 Such innovation has always been a part of the process at DEC, says O'Loughlin.

 The PDP-11 represented a totally new computer architecture. It required new peripherals, new system software, new diagnostic software, new manufacturing procedures. The UNBUS made it possible to introduce the PDP-11/20 with a minimal hardware configuration. Options could be added later.

 Introduction was scheduled for early January 1970, and the company still had to gear up in dozens of ways to get ready to make market and service this new line of products.

 In late August 1969, Julius Marcus joined the company as PDP-11 marketing manager. "I had been with the company only two days when I was told to go before the Operations Committee with a proposal to price the PDP-11," he remembers. "I had just a week to come up with a price list and documentation to justify it relative to the competition."

 In September, Tom MacDonald came on board as the production engineer responsible for introducing the PDP-11 into manufacturing.  Almost immediately, before the final designs had been released, he had to make commitments for shipments that were a year away -- otherwise materials, equipment and people would not be in place when they were needed.

PDP-11 patents

 Although DEC, in 1969, was not very patent-conscious, Andy Knowles convince the company to file for patents on the PDP-11 because of the computer's unique design and market potential.

 Four basic patents cover the whole PDP-11 family, including the LSI-11. (The company has other patents on specific models, but these four apply to all models.)

 Processor Design (two): 1. Hal McFarland and Jim O'Loughlin, investors; 2. Bruce Delagi, Hal McFarland and Jim O'Loughlin, inventors

 UNIBUS and UNIBUS Peripherals (two): John Cohen, Paul Janson, Hal McFarland, and James (Bev) Young, inventors for both.

Debut in January

 Engineering, marketing and diagnostics were all on the same floor, Mill 5-2. Field service was intimately involved, with Andy Verostic, in particular, not only helping debug prototypes but also contributing to the design. Systems software was in close touch, suggesting -- and getting -- changes in the hardware design.

 Although the product had tremendous potential for expansion, it would be a year or two, at best before a full complement of options would be available. In the meantime, it would have to be sold with just 4K or 8K words of core memory, some rudimentary paper tape, software, a Teletype, a paper tape reader and punch. And, it would be in competition with more elaborate systems already in production for a year or more.

 "The challenge," according to Marcus, "was to make this product look unique."

 The potential of this new family of computers was emphasized -- the UNIBUS and the new architecture with its large address size, easy interfacing and easy programming.

 The original press release, sent out Jan. 2, 1970, boasted that because of the UNIBUS, the PDP-11 would change the concept of computer obsolescence: "...the UNIBUS offers a customer a new approach to computer expandability as well as extending computer life almost indefinitely."

 The sales force considered the PDP-11 a very complicated computer. To attack this problem, Marcus did two things. He sent Jim Davis to the regions and larger districts to give sales representatives two-day courses on the PDP-11. He also asked for 10 hand-picked salesmen, at least one for every region, and brought them in for special training. NO equipment was available at the time, so he had to make do with audiovisuals.

 "As PDP-11 sales specialist of the Central Region, I was like a roving ambassador," Gale Morgan remembers. "I would be called in to make a customer presentation and then go on my merry way."

 Morgan, now Western Region manager, made the first quantity sale -- 25 machines to Applied Dynamics in Ann Arbor, Mich. "I remember talking about the 'family of machines.' Every time I'd say that, somebody would ask, 'Well, what are the next machines going to look like?" I had no idea. 'Are they going to be bigger or smaller?' 'I don't know, but it's going to be a great family of machines.' Every time somebody would bump e, I'd say, 'great family of machines.'"

 Dezi Dezzani who made one of the first sales to MB Electronics in New Haven, Conn., recalls: "We were selling against real hardware, and all we had to show was paper. The customer had to have a great deal of confidence in DEC and our ability to come through," and many did.

 By the end of the first week after the introduction, the company received 150 orders for PDP-11/20s. By March, the product line had booked its first two $1 million orders -- 100 computers for CMC in Canada and 10 systems for Sundstrand in Chicago.

 In those early days, engineers frequently went along to help on sales calls. "Some customers thought the PDP-11/20 was slow because the instructions cycle time was slow," said Tom Stockebrand, engineering manager for communications. "But the instructions were powerful enough that you needed lots fewer of them. You had to teach that to customers."

 In spring 1970, Stockebrand came over from process engineering to design communications products for the PDP-11. His group, including Vince Bastiani, quickly developed a series of modules that interfaced between the UNIBUS and telephone equipment.

 Such hardware and corresponding software helped open up a whole new marketplace in distributed networks and communications, Cady said.

 By April, when the first two systems were delivered to customers, dozens of PDP-11 development projects were under way. From May until November that year, the company announced 17 new products related to the PDP-11. Options were being released so fast that price lists were incomplete by the time they were printed.

 Demand for information was so great that system manuals were released in rough form to get them out early and published later in final form.

 "We thought 10,000 PDP-11/20 handbooks would take care of us forever," Marcus recalled. "In just two or three years, we distrusted over 300,000."

Teamwork makes it possible

 Manufacturing had been supporting engineering since early fall, and engineering, in turn, supported manufacturing, said Tom MacDonald, senior production engineer. Well past the initial shipments of systems, the two groups worked together as a single project team.

 To save time on the first two shipments, manufacturing built the boards and the sheet metal, and engineering assembled and debugged the systems. Meanwhile Ron LaFosse set up a pilot production line in Mill 1-5.

 Field Service technicians who would have to install and service early shipments got on-the-job training by helping manufacturing gear up for production. The Westfield plant, where the production line would ultimately move, also sent representatives to Maynard for training.

 In June, the first month of production, 97 computers were shipped -- "within three of the 100-system schedule established over six months ago," noted Bob Puffer, in a memo commemorating the event. In July, over $1 million worth of PDP-11s were shipped.

 By Nov. 3, the entire production line had moved to the newly-opened Westfield facility. Even then the product line stayed in close touch with manufacturing, through monthly steering committee meetings in Westfield.

Family adds new members

 Also that fall, the family's first operating system DOS (Disk Operating System) became available, making it possible to use peripherals besides paper tape (e.g., DECtape, a fixed head disk or a removable disk).

 The original software had consisted of just a few paper tape utilities -- an assembler, a loader, an editor and a rudimentary operating executive. Since prototypes were not available, programmers had to use a simulation on a PDP-10.

 "We weren't able to do much work," system programmer Roger Willis recalls. "The simulator was written in FORTRAN an used so much of the large machine's resources that only one programmer could use it at a time."

 By the summer of 1970, the DOS development team had its own PDP-11/20 but still needed to use the simulator on the PDP-10. The tam offices were in Mill 12-2, the computer lab was in the lobby of Mill 12-1 and the PDP-10 in the basement of Mill 5, according to another systems programmer, Louis Cohen.

 "We would work at our desks, figure out what changes we wanted to make in the program, run downstairs to the PDP-10 terminals to edit files, run up to Mill 12-3, all the way across Mill 3 to Mill 5, down to the basement to get the listing and the paper tape; then come all the way back to our desks, figure out what we had done, go back downstairs and try the paper tape on the PDP=11/20," Cohen said.

 A year after the product introduction, the U.S. economy slipped into a recession, but the PDP-11 was strong. "We would not have been as successful as we were if we had been a year later. Timing was critical," said Cady. "As it turned out, we had a backlog and heavy demand for a unique product during that recession. We kept moving."

 Not only did the PDP-11/20 sell, but customers kept ordering larger and more expensive systems than anticipated.

 The product line had expected heavy demand for inexpensive basic units -- PEP-11/20s with 4K words of memory and an even small computer announced at the same time -- a dedicated controller called the PDP11/10. As it turned out, customers ordered large rather than small PDP-11/20s and the PDP-11/10 was never shipped. (Two years later, another PDP-11/10 -- entirely distinct from the first -- was announced.)

 "I remember we used to have raging arguments over what the average size of a system would be," said Pete Durant. "Some people said, "You'll never need more than 4K words of memory," Others said, "We're going to go to 8K someday." The first year, the average system was over 8K."

 Because of the expected heavy sales of small, basic computers, systems programming had squeezed DOS into a very small space -- just 1K of memory, John Sosville remembers. the effort required tradeoffs, ingenuity and lots of time.

 "It was like packing all you need for a vacation into a briefcase only to find out later that you could have used a trunk," Sosville said.

 Software development manager Hank Spencer continued: "When systems began to et much bigger and memory prices began to drop, there were inefficiencies inherent in DOS that we could not get around anymore. It would have been considerably more efficient had we designed for a larger machine int he first place.:

 As it turned out, PDP-11 business simply did not follow the pattern of PDP-8 business.  Customers bought larger systems for larger jobs. In fact, the PDP-11 did not simply replace the PDP-8, as many had expected it would, but rather opened new markets, new applications. In terms of units shipped, the PDP-8 business continued and still continues to grow, notes Marcus.

 In 1970, a basic PDP-11/20 would fill about half a bay, but within a couple of years, customers were ordering systems with six or seven bays of peripherals.

 "We were surprised. Customers recognized the value of the machine much faster than we expected," admits Marcus. "We were selling the PDP-11 as a simple product and customers were buying it as a complex product. That led us to change direction. We had to learn to engineer, build and service large computer machines."

 Focus shifts to applications

 "We believed the market was only limited by our ability to do a good job for our customers," Andy Knowles said. "We felt we could grow at least 40% to 50% a year for at least five years. We set plans for bookings and shipments accordingly. But the PDP-11 exceeded every plan that we ever made."

 After DOS, which was a general purpose operating system, came RSX-11D and RSX-11M for real time applications; RSTS and RSTS/E for timesharing and RT-11, a small single-user system for program development.

 Meanwhile, the computer evolved in three directions -- toward less expensive systems with about the same performance; toward high performance systems at about the same price and toward high-priced maximum performance systems.

 Each generation of equipment took full advantage of advances in technology.  For instance, multilayer boards were used int he PDP-11/45 and large scale integration int he LSI-11 -- maintaining software compatibility as much as possible.

 Original sales wee mostly to highly technical customers. then, as the software became richer and the range of computers and options expanded, the PDP-11 "went all over and into everything," according to Julius Marcus, now Vice President, Commercial Group.

 "The market was hungry for a 16-bit product from the leader in minicomputers. The market was ready for it and it took off rapidly."

 The phenomenal growth of the product led to changes in the corporate structure. Product lines shifted from a product focus to an applications and market focus. PDP-11 engineering became part of the new Central Engineering organization.

 "The PDP-11 went from being something that was done in a corner to a pervasive, all-inclusive activity that gradually absorbed the lion's share of the company's interest," Marcus said.

 The company, in 1979, introduced two new, fourth generation members of the family -- PDP-11/23 and PDP-11/44. DEC facilities from Albuquerque, N.M. to Galway, Ireland are engaged in PDP-11 manufacturing. The PDP-11/03, /04, /23, /34,44 and /60 central processors (CPIs) are made in Aguadilla, Puerto Rico; PDP-11/70 CPUs in Burlington, Vt. and Galway, Ireland; and LSI-11s in Albuquerque, N.M. (In May Albuquerque will celebrate shipment of their 100,000th COU. Soon LSI-11/23 SPUs, now made in Westboro, will be transferred to Albuquerque.)

 Over the last ten years, DEC has manufactured over 200,000 16-bit CPUs, including PDP-11 computers and LSI-11 boards. In the words of Andy Knowles, now Vice President, Technical products: "The PDP-11 is by far our biggest product."

Bruce Delagi

	Webseltzerbooks.com