Departmental Computing Centers

Minicomputers

During the 1970s, the economies of scale in computer hardware were gradually reversed. I do not understand the engineering reasons that caused this, but it coincided with the development of large scale integrated (LSI) semiconductors for processors and the phasing out of magnetic core memory. A new type of computer emerged, known as minicomputers, and a new group of computer companies. The market leader was Digital Equipment Corporation. By the late 1970s, minicomputers such as Digital's PDP 11/70 for timesharing and VAX 11/780 for number crunching were as cost-effective as the large central computers, and for some tasks were clearly superior

From the 1960s IBM had dominated mainframe computing in the United States, though other companies did well in certain markets, such as Burroughs in banking, and Univac and Control Data Corporation in scientific computing. The mainframe companies were popularly known as "IBM and the Seven Dwarfs" (Burroughs, Univac, NCR, Control Data Corporation, General Electric, RCA, and Honeywell). None of them foresaw the emergence of minicomputers and a new group of companies replaced them. For universities, Digital was the most important minicomputer company, but other companies such as Data General, Hewlett Packard, Wang, and Prime were also successful. Data General's Nova was popular for computer graphics, and Wang was the leader in word processing and office automation. Many of these companies were spin-offs from MIT and scattered around Route 128 outside Boston.

For most of the 1970s the most widely used family of minicomputers was the Digital PDP 11. The smaller members of the family were used to control laboratory equipment and the larger ones could run a substantial timeshared system. At the end of the decade, 32-bit minicomputers such as Digital's VAX 11/780, the Data General Eclipse, and the Prime 750 had completely reversed the economies of scale. At Dartmouth, we had a Prime 750 for medium scale number crunching and a variety of VAX and Prime computers for specific applications, such as the library's online catalog and the alumni database. At larger universities, research groups set up their own departmental computing centers. The first book to popularize the high-paced culture of computing was "The Soul of a New Machine" by Tracy Kidder, which described the rush to bring the Eclipse to market in 1980.

These minicomputers computers had excellent operating systems. While IBM's MVS operating system grew bigger and bigger until it became a nightmare to install and upgrade, the Digital and Prime systems were comparatively straightforward to manage. At Dartmouth we chose Prime because they supported the PL/1 programming language and the X.25 networking protocol, both of which we used for DTSS, but Digital's VAX with its VMS operating system proved more successful in the long term. Several members of the VMS group later moved to Microsoft where they were influential in developing the Windows operating systems.

Departmental computers

Minicomputers had a profound impact of the organization of university computing. For years, the management of central computers had assumed that hardware was a critical resource, particularly central processing cycles. To waste a single cycle was a crime. Only gradually did we come to realize that the time of the faculty and students was much more valuable. When I was on sabbatical at Dartmouth, I watched a group of mathematics professors who kept trying the same iteration, with ever-longer run times. Eventually they realized that they were iterating over a singular matrix and it would never converge. I was horrified at the waste of computer time until I realized that they had gained mathematical insights that were worth much more than the computer time. Not many years later I myself ran an integer programming problem for an entire weekend on a Prime 750. It would have been irresponsible to do so on a shared computer, but not on a machine that would have otherwise been idle.

When a university had a single central computer, the hardware determined the organization. A large central computer needed a team to manage it and to share its resources among departments and users. Almost every university created a computing center with its director. Most of these centers served both academic and administrative users. The center would have a system programming staff to support the central machine and applications programmers for administrative computing.

The centers never fully solved the problem that different groups of users have different computing needs. Most academic users run large numbers of small jobs, but some people want to run big computations or process very large sets of data. Administrative computing has an entirely different set of needs. In aggregate the capacity of the central computer was never enough to satisfy everybody. With varying success, computing centers attempted to balance priorities by technical, administrative, and financial mechanisms, but they could not make everybody happy. For researchers with research grants, this unhappiness was aggravated by a peculiarity of how many universities charged for computer time.

Research universities wanted to recover the cost of research computing from funding agencies, such as the National Science Foundation. These universities charged all users for machine time. Researchers used their grants to cover the costs, while other departments paid from their own budgets. To recoup as much money as possible from grants the universities set their computing charges at the highest rate that the government would allow, including full overhead recovery.

Frustrated by these high charges and the inflexibility of central computing, the richer departments used their research grants to buy minicomputers and set up their own computing centers. Staff costs were largely eliminated by having graduate students look after the systems. When I arrived at Carnegie Mellon in 1985 the university claimed to have more VAXes than classrooms. There were well-run centers in computer science, electrical engineering, physics, statistics, and several other departments. Later, as personal computers became widely available, schools and colleges, such as fine arts and the business school, set up personal computing centers.

Many computer directors felt threatened by these developments. Some of these feelings were justified. The cost comparisons were often dubious, since departments did not include the indirect costs of running a computer center which are substantial. The typical departmental computer was run by an assistant professor and a graduate student. Only too often the student never graduated and the assistant professor did not get tenure, but the researchers clearly saw departmental centers as an effective way to spend their resources. An engineering professor at Dartmouth explained to me the advantages of controlling his own computing, free from the juggling act that is inevitable with a central computer that serves the entire university. Recently, for my research at Cornell, I was in a similar position. Our group had its own large computer cluster and we made all the decisions about how to use it.

Computer centers were also worried about the costs of supporting large numbers of different types of computers. Undoubtedly such proliferation did cause difficulties and universities tried various approaches to limit the variety. When I arrived at Carnegie Mellon there were 101 different makes of computer on campus. One of my first acts was to abolish an ineffective policy where I, as vice president for computing, had to approve all purchases of departmental computers. In later years we developed a short list of personal computers that we supported centrally. Most members of the university bought these types of computer, relying on us for support, but there were always a few mavericks who chose differently. These mavericks were invaluable in evaluating new options.

At the universities that I know best, Dartmouth, Carnegie Mellon, and Cornell, the various organizations eventually learned how to work together and support each other, but even today there remains an underlying tension between centralization and decentralization at almost every research university.