The Ancient History of System/360

On April 7, 1964, the International Business Machines Corporation changed the business world with one of the most momentous gambles ever made by a corporation. It introduced the System/360 family of computers.

Despite the staggering evolutionary leaps computers have made in the last quarter-century, fundamental technical concepts introduced with the 360 are still part of the backbone of the general-purpose computer industry. System/360 was the first product family that allowed business data-processing operations to grow from the smallest machine to the largest without the enormous expense of rewriting vital programs. Before then even IBM computers were a hodgepodge of incompatible machines. The few commercial computer users around faced major conversion problems and upheavals whenever their businesses grew too large for one machine and they were forced to switch to another—even from the same manufacturer. Small, medium, and large machines required their own armies of operators, even though they were being built to solve similar business problems.

System/360 introduced upward and downward compatibility, allowing businesses for the first time to grow and prosper without being penalized. The system’s popularity brought it highvolume sales, which in turn drove down manufacturing costs. This was a turning point in beginning the great drop in the cost of computing that has accompanied the exponential growth in computers’ power.

Without technical concepts first embedded in System/360, the United States might still have not landed men on the moon. Automobiles would be less safe. Telephone systems might be much more complex and harder to use. Transportation and distribution networks for factory and farm products might be far less efficient.

As with any project of such magnitude, it was far from the work of one person. Facing stiff competition from companies like Remington Rand, General Electric, and Honeywell as the 1960s began, IBM’s top management risked almost everything, investing five billion dollars on a new line of computers that would make all the existing ones obsolete. “We call this project ‘you bet your company,’” one executive told Fortune magazine. But if there was one person most responsible for making the wager pay, for driving System/360 through the corporate technical and political bureaucracy from initial conception on, that person must be Dr. Robert O. Evans.

Born and reared in a small Nebraska town and trained in electrical engineering at Iowa State University, Evans joined IBM in 1951 as a junior engineer on one of the company’s first computer projects, the Defense Calculator, later known as the 701. He helped run what he calls “sort of a nationwide emergency service” to keep the machines operating and customers happy. In 1955 he was noticed by Jerrier A. Haddad, an IBM executive in charge of new products. Haddad made Evans his administrative assistant, and Evans proved his worth when problems arose with a system built by IBM for the National Security Agency. Evans handpicked a group that worked seven days a week for nearly eight months to rebuild the NSA system and get it working.

An imposing, barrel-chested man, Evans was unyielding when he be lieved he was right. And because he usually was right, he earned top management’s confidence and the power to pull strings within IBM to make things happen his way. He worked tirelessly, including on weekends. When making a point in a meeting, he might pound his fist on the table, but those who knew him best saw a gentle man behind the tough exterior—a man admired by those who worked for him and by top management too.

After fixing the National Security Agency’s problem, Evans was promoted to run a top-secret IBM department serving the Navy. Later, in 1958, he was put in charge of all processing systems under development by the newly formed General Product Division (GPD), in Endicott, New York. GPD was one of two units within IBM that designed computers. The other—GPD’s big rival—was the Data Systems Division (DSD), in Poughkeepsie, New York. GPD was responsible for lower-priced machines, DSD for bigger, higher-priced mainframes. And DSD had the task of planning the next generation of computers.

In January 1961 DSD was on the verge of unveiling a new line of computers, the 8000 series, that was to replace many of its existing computers. The 8000 was well received in presentations to corporate top management. However, as Frederick Brooks, who was in charge of the project, recalls, at one showing there was “one person who was extremely distressed by what he saw—and stood out like a dark cloud in the back of the auditorium.” That person was T. Vincent Learson, the corporate executive in charge of both DSD and GPD and later the chairman of IBM.

As Learson noticed, the 8000 was falling short of expectations. Significant improvements had been planned in circuit technology, but pressure from the sales force to get the machine out fast had led to a decision to stick with older technology. And there were no plans to make machines within the 8000 series compatible with one another. Some were to be organized using words, others with characters. Instruction sets would differ.

At the same time, other incompatible computers were continuing to be developed by GPD. Altogether seven noncompatible families of IBM computers had emerged in the preceding ten years. As Evans later recalled, “With so many types of architectures, IBM was spending most of its development resources propagating the wide variety of central processors. Little development effort was devoted to either peripherals or programming. A user could move from one processor in a family to one twice as fast, but only achieve 10 percent improvement in throughput [problem-solving ability], since the existing disk or tape peripheral devices and programming could not keep pace with the central processors.”

Learson wanted to change this situation. Virtually overnight he thrust Evans into the DSD organizational hierarchy, below Charles DeCarlo, a DSD vice-president, but above Brooks, DSD’s crack computer designer. Learson asked Evans to take a close look at whether the 8000 series should be announced at all. “His words were essentially, ‘If it’s right, build it, and if it’s not right, do what’s right,’” Evans says.

It was the first encounter between Evans and Brooks. Brooks, born and bred in North Carolina, had become enthralled as a teenager with the earliest computers. After graduating from Duke University as a physics major, Brooks went to Harvard to study under the computer pioneer Howard H. Aiken, and he eventually received a Ph.D. in computer science there.

From Harvard Brooks joined IBM in 1956 as an associate engineer and, like Evans, moved up fast in the technical ranks. He started out in a group designing a machine called Stretch that was supposed to stretch existing technologies. Stretch failed to meet its goals and caused IBM considerable embarrassment, but Brooks was given more key assignments. He helped design a computer for a critical classified government project and then, in 1960, worked on an interdivisional task force developing what was known as the May Day Machine. After that Brooks was promoted to systems manager of an unannounced computer line being developed in Poughkeepsie. That was the 8000 series.

When Evans became his boss, he recalls, “one of Bob’s first orders of business was that he and I go out to supper together. He tried to explain to me why the 8000 series was the wrong thing for the company to do. I did not find his arguments convincing.” Brooks, like Evans, was persistent when he believed he was right. Unlike Evans, he was usually soft-spoken. For several months a fierce internal political battle raged over the fate of the 8000 project—with Brooks and his supporters on one side and Brooks’s new boss, Evans, on the other.

In the January 1983 issue of Annals of the History of Computing , Brooks recalled both the war and the truce that ended it: “Bob and I fought bitterly with two separate armies against and for the 8000 series. He was arguing that we ought not to do a new product plan for the upper half of the business, but for the total business. I was arguing that that was a put-off, and it would mean delaying at least two years. The battle … went to the Corporate Management Committee twice. We won the first time, and they won the second time—and Bob was right.”

As Evans bluntly remembers it, “I thought [the 8000 series] was poor, and I also thought the technology was a bum choice, and we could do a lot better.”

By April Evans’s superior at DSD, Charles DeCarlo, had been replaced by Evans’s old boss Jerrier Haddad, who supported Evans’s urgings for a better product line. Evans, winning the battle to kill the 8000 but realizing he had many wounded egos to heal, took all his adversaries, including Brooks, “away to the Gideon Putnam, a hotel in Saratoga Springs, to look at our belly button and decide how we were going to organize and what we were going to do.” Evans allowed some Poughkeepsie favorite-son projects to continue, but his heart and mind were set on the New Product Line (NPL) concept of total system compatibility. He made a dramatic overture: “I so respected Brooks for his intellectual ability, I surprised him by offering him the senior job on what became System/360”—the job of systems manager. “He surprised me by taking it.”

“To my utter amazement,” Brooks recalls, “Bob asked me to take charge of that job after we had been fighting for months. 1 was dumbstruck at being asked to take charge of the juiciest part of his work.”

Evans’s victory in getting NPL started wasn’t complete yet, and to cement it he would need the help of Donald T. Spaulding, a former boss who now led Learson’s Group Staff—a half-dozen executives who could cut through organizational lines and strongly influence future product decisions. As Evans recalls, “Spaulding was very concerned about a key guy named John Haanstra.”

Haanstra was with GPD, which was producing one of IBM’s most popular and profitable machines, a small processor known as the 1401. The 1401 was threatened by a lower-priced competitor, the Honeywell 200, which came with a software tool that permitted it to run 1401 programs. The machine clearly worried Haanstra, and Evans feared he might bolt from supporting the New Product Line and muster all his division’s energies behind a 1401 follow-on product. This would be a serious blow, since the low end of a typical computer product line has the highest volume of sales and actually produces the most profit. Therefore, the low end, for which in this case Haanstra’s group would be responsible, would make the upper end feasible. Without GPD’s support, System/360 would never see light.

This was especially true because System/360 was going to rely on new solid logic technology (SLT), a hybrid microminiature circuit design and packaging technology that would allow automated manufacturing on a mass production line. Without the low end the volume of components being made would be too small to make economic sense.

Haanstra “was making sounds as if he would support the New Product Line, with [GPD] doing the low end,” Evans says, but “Spaulding didn’t really believe him, because the able John Haanstra had demonstrated some erraticism in the past.” So in October 1961 Spaulding proposed an international top-secret task force to plan every detail of the New Product Line and make necessary compromises along the way. And in a stroke of political genius, he named as chairman of the group none other than John Haanstra. To no one’s surprise, the vice-chairman was Bob Evans. In classic IBM obfuscation the task force of thirteen men—primarily top technical experts representing all the company’s manufacturing and marketing divisions—was called the SPREAD (Systems Programming Research and Development) committee. Evans and other insiders preferred to say SPREAD meant “Spaulding’s Plan to Reorganize Each and All Divisions.”

Throughout November and December 1961 the thirteen met daily in strictest secrecy at the Sheraton New Englander Motel, in Greenwich, Connecticut. Their lengthy sessions were punctuated by heated technical arguments. When they concluded, on December 28, they had hammered out the framework for the System/360 product line, a detailed technical document dubbed the SPREAD report.

As Evans candidly recalled, “the stated purpose of the group was to develop a common plan. … But the real purpose was to get John Haanstra’s name on the line, signed up in front of the corporation, to do the low end. The plan went a bit astray because, at the end of November [1961], Haanstra was promoted and went back to a bigger job. I took over as chairman. … Spaulding and I strategized, and we had Haanstra make the report to senior management in January 1962.”

Joining Frederick Brooks as a chief designer of the system was another young computer architect, Dr. Gene M. Amdahl. Amdahl was from South Dakota, where electrification did not reach his parents’ farm until he was in high school. Before that he had tried to make a cat’s-whisker radio, but, he said, “there were no stations near enough to ever find out if it worked. You could get a little static, but that’s all.” He had earned his Ph.D. in theoretical physics at the University of Wisconsin. Evans saw to it that Amdahl played a key role in designing System/360’s architecture. He explains: “Amdahl, in days before there were truly computer architects, was … a brilliant architect. I have yet to see his peer. He could visualize what happens internally in a computer during the computational process … and the flow of things during the solutions of problems. Amdahl kept the designs reasonably simple. He had a very good view of how data moved in a machine, and 1 thought that, in designing the data paths and designing the instruction repertoire of this family, Amdahl would be invaluable.”

Amdahl, for his part, speaks of Evans as “a master at manipulating people. I liked him. I always got along well with him. But I always knew when he was manipulating me … and I would act in such a way as to make it come out the way I wanted it.”

As Evans recalls, there was terrific feuding between Amdahl and Brooks while the two star designers worked on the 360 project together: “There was a time in 1963 when Brooks and Amdahl decided they couldn’t work together anymore. They came to me, and I had to ‘shoot’ one or the other, so I concluded I had to shoot Brooks, as I believed the project couldn’t afford to lose Amdahl. So I shot Brooks. He left the project.”

Then the president of DSD, W. B. (“Bill”) McWhirter, who “was out in Aspen at an executive conference, came running back and interceded with Brooks and pleaded with him to come back on the project,” Evans says. “And he did. I accepted him back, and we went on.”

It appears that while Evans was the chief proponent of upward compatibility, making the basic 360 machines compatible with the largest mainframe machines, Amdahl had much to do with extending compatibility downward to encompass the smallest computers.

Evans, when asked where the upward-downward compatibility notion originated, says, “I believe it started primarily with me,” but adds, “the real seeds started years earlier when [IBM’s] new San Jose laboratory had built a disk-based system called the 305,” which was to be compatible with at least one other IBM processor. Amdahl, however, says downward compatibility was his idea: “I told them I wouldn’t take the job unless I could do that.” In a commentary in the April 1984 Annals , he wrote: “When I was being recruited by Bob Evans to head the architectural group, the task was to define a line of upward compatible processors. … I told Evans I’d do it, but insisted that the line of processors must be upward and downward compatible, or the same chaos would ensue.”

Evans agrees that “the downward was really the trick. The issue was to have a machine that’s skinny enough that you can profitably produce the low-cost, high-volume entry at the bottom of the line, and whether the constraints that are put on fulfilling the bottom of the line give enough performance at the high end. So that was the problem as we went to work.”

In early 1962 Brooks held a competition among thirteen IBM engineering teams to come up with design ideas for the system. The winner was Gerrit A. (“Jerry”) Blaauw, who, along with Amdahl, had the crucial idea of using base registers. A register is a storage device having a specified capacity, such as a bit (a single binary digit—the smallest unit of information), a byte (a sequence of bits), or a computer word. Blaauw and Amdahl’s idea sped the processing by using abbreviated addressing, allowing data to be retrieved from storage by using shortened codes. Because shorter codes would require less memory, this would give greater power to small machines, and because of that, according to Brooks, it “was the key step that got us over the compatibility hump—and they had independently come up with that. It was not a totally new idea in the field, but it was new in the IBM company.”

Brooks says the reason Blaauw edged out Amdahl in the competition was that “Jerry’s machine was based on sizes that were powers of two and Gene’s were based on sizes that were three times powers of two. So [Amdahl’s] were twentyfour, forty-eight, ninety-six, with three index bits and everything that follows from that, and the six-bit byte. Jerry’s were four, eight, sixteen, thirty-two, sixty-four, with an eight-bit byte. That led to a great ‘six-eight’ fight that took place in June.” The eight-bit byte ultimately won because, having more permutations, it made possible a wider range of alphanumeric characters and symbols. Eight bits allowed 256 different combinations, meaning System/360 could be used for a wider range of business applications and also, eventually, in other countries around the world that had different alphabets and symbols. The eight-bit byte has been an almost universal computer standard ever since.

Despite the hidden motive behind the SPREAD meetings, the SPREAD report stood the test of time. During the next two and a half years, as IBM engineers and scientists labored irr secrecy to build System/360, the SPREAD recommendations were followed without any significant deviation. The technology the report proposed not only became embodied in the bowels of System/360 but set many technical standards that are still followed by IBM and the rest of the industry today—and will probably continue to be well into the 1990s.

Evans listed seven basic points of the SPREAD report in an article in the January 1983 Annals . First, the central processing units were to be both versatile—handling scientific and business applications with equal ease—and totally compatible with one another. Second, they all were to accept the same peripheral devices, such as disks and printers. Third, although integrated circuits were being considered at the time, the processors were to rely on hybrid microminiaturization, chips that had discrete components wired on—a less advanced but also less risky technology. Fourth, uniform programming would be developed for the whole line. Fifth, a single high-level language that could serve both scientific and business needs would be designed. Sixth, the processors would have the power to address an unprecedented sixteen million characters, expandable to two billion characters. Seventh, the basic unit of information would be the eightbit byte. In conclusion Evans wrote: “Perhaps most important, the SPREAD report permitted IBM to focus on an excellence not possible with multiple architectures. It resulted in powerful new peripherals, programming, terminals, high-volume applications, and complementary diversifications whose future can only be imagined.”

Dr. Erich Bloch, who retired from IBM in 1984 to become director of the National Science Foundation, managed the development of the solid logic technology. By successfully unifying SLT circuit and package technology across the entire product line, he eased the way for broad standardization, which not only lowered manufacturing costs but also simplified training and reduced the number of parts needed at service centers. “Without [Bloch], we wouldn’t have been in the game at all,” Brooks has said.

Although System/360’s hardware was in reasonably good shape by the time of the April 7, 1964, public announcement of the line, its operating system, known as OS/360, was not. An operating-system plan had been part of the project from the time of the SPREAD report. However, at the end of 1963 the programming group responsible for it went to management with a plan known as the Romans because it proposed four operating systems identified by the Roman numerals I, II, III, and IV.

“There was little compatibility among the Romans,” Brooks says. They were essentially four independent programming systems for different sizes of machines, completely defeating the compatibility goal. The Romans plan was killed, and a team of experts was sent off to a retreat in the woods in February 1964 to come up with a better idea. The group laid the groundwork for OS/360.

However, that system had problems too. It turned out to be the most complex operating system IBM had ever produced, a far more difficult design task than anyone had expected. As of the summer of 1964, several months after System/360 had been officially announced, the operating system was still full of bugs. At Evans’s request, Brooks took over the matter. He hoped to get it cleared up by August, when, he told Evans, he planned to resign from IBM to head the computer science department at the University of North Carolina. Thomas Watson, Jr., the chairman of IBM, personally intervened and talked Brooks into working on the project one more year. In March 1965 the operating system successfully passed “Alpha test,” which, Brooks said, “was the objective I was determined to see done when I stayed the extra year.”

Those who succeeded Brooks wrestled with additional performance problems into yet another year, 1966, to the intense dismay of the IBM marketing people. There was one “big miss on 360,” in Evans’s words: its planners’ failure to adopt a concept called dynamic address translation, which would have made time sharing much easier. Time sharing, which was very important in business computing until the personal-computer era began, permitted multiple remote users to use the same computer, via terminals, at the same time. Dynamic address translation, later also called virtual memory, automatically juggled users’ data and programs, moving them in turn between main memory and disk storage.

The Massachusetts Institute of Technology was a major proponent of dynamic address translation. IBM’s chairman Watson had instructed Evans to track what was going on at MIT. Evans sent experts, including Brooks and Amdahl, there, but the team concluded that changing the architecture to incorporate dynamic address translation would take too long and cost too much. A competitor, General Electric, listened to MIT, and one result was that the prestigious university purchased a GE system instead of one from IBM. Another leading-edge account, Bell Laboratories, shortly afterward did the same. But System/360 was so broadly successful elsewhere that it helped put GE out of the computer business entirely by 1970.

The first 360 computer released for sale—the 360-40—started being shipped in April 1965. Three other models soon followed. IBM’s marketing staff breathed a huge collective sigh of relief. Honeywell had already unveiled its 6600 computer, designed to compete with the 360, and an exasperated Watson had sent around a memo complaining that Honeywell had developed its machine faster in a research laboratory of only thirty-four people—“including the janitor.”

But IBM had more orders waiting for the 360 than it could handle. In 1966 the company hired twenty-five thousand people worldwide and began building three million square feet of manufacturing space. By the end of that year a thousand 36Os were being produced each month. The financial records speak for the smashing ultimate success of IBM’s gamble. In the six years from 1966 through 1971, IBM’s gross income more than doubled, from $3.6 billion to $8.3 billion. Net earnings after taxes did the same, rising from $477 million to $1.1 billion. In 1982 the direct descendants of System/360 were still accounting for more than half of all IBM’s gross income and earnings.

IBM’s primary computer architecture today is System/370—an evolutionary, not revolutionary, step beyond System/ 360, with many of 360’s concepts embedded in it. System/360 and 370 architecture has been adopted by hundreds of companies, including many Japanese manufacturers—who make plugcompatible peripheral devices and processors—and thousands of software firms selling control and application programs. All this has helped prolong 360’s mortality, as has the development of major operating systems using its architecture, many of which have themselves become industry standards: COBOL, the most widely used business data processing language; FORTRAN, a language used by scientists and engineers; and the Disk Operating System (DOS), the basis of the personal computer.

Most of the men responsible for making System/360 happen have thrived too. Robert O. Evans held a number of executive positions at IBM, retired in 1984, and is a general partner in a San Francisco-based investment firm. Frederick Brooks left IBM in 1965 to head the department of computer science at the University of North Carolina, where he remains active today. Gene M. Amdahl resigned from IBM in 1970 and founded an IBM archrival, the Amdahl Corporation, which produced a line of IBM-compatible processors. He has since left Amdahl and begun two other start-ups. T. Vincent Learson became chairman of the board of IBM when Watson retired in 1971 and is now retired himself.

Few of the System/360 computers themselves are still active. Most have been obsolete for years. But the dominance they gave IBM in the business computer design field is a legacy “that has endured, however challenged, through many computer generations.