Some Burroughs Transistor Computers
Unisys History Newsletter.
Volume 3, Number 1
by George Gray
The Burroughs experience with transistor computers was in many respects parallel to that of Sperry Rand. At Burroughs too, early development of transistor computers for the military was not carried over in a timely manner into the large-scale commercial market, but the machine which emerged, the B5000, was a very innovative design. As was the case at Sperry Rand, Burroughs small computers kept the company in the computer business during the transition from vacuum tubes to transistors.
The Atlas Guidance Computer
The Burroughs Great Valley Research Laboratory at Paoli outside Philadelphia had from its beginning been heavily involved in military projects. It developed the AN/FST-2 data transmitting set for the Air Force's extensive North America air defense system which was deployed during the 1950s. When the system was complete, 134 of these data communications devices had been installed. The Air Force also turned to Burroughs for a transistor ground guidance computer for the Atlas intercontinental ballistic missile (ICBM). It was officially designated the AN/GSQ-33, but was usually referred to as the Atlas computer. The design was done by Isaac L. Auerbach, a former employee of Eckert-Mauchly who had joined Burroughs in the late 1940s. Work on the project began in April 1955, but problems with the transistors supplied by Philco slowed things down. In a 1992 interview, Auerbach recalled: "We had a major problem there. Philco was going to provide the surface barrier transistors. One of the conditions in the contract was that they could not move the laboratory line where they were manufacturing these until their production line was up running and we approved the quality of the transistors coming off the production line. Philco, without talking to us, set up a production line, cut it over without going parallel, and we got a pile of junk coming in." The problems showed up in testing, and Auerbach called Philco: "Look guys, I don't care what you're planning to do. You set that lab line back up again and you run them on that line, and if you have to, you run the entire production line that we want on that line. I'm not whistling. Here's the project, here's the priority, don't screw around with this". (Oral history interview of Isaac L. Auerbach, October 2-3, 1992, conducted by Bruce H. Bruemmer ,OH 241, Charles Babbage Institute, University of Minnesota) Eventually the problems were worked out, and the first machine was installed at the Cape Canaveral missile range in June 1957. Although Atlas missile launches started in September 1957, test patterns were transmitted to the missile in place of actual guidance commands for the first four flights. The first computer-controlled launch was on July 19, 1958. The computer had separate memory areas for instructions (2048 18-bit words) and data (256 24-bit words). The instruction area was increased to 2816 words, beginning with the Model III version, which was first delivered in December 1958. The Atlas guidance computer had no facilities for developing programs, so they were written on the UDEC II, the Datatron, and the 220, using simulator software. Burroughs was still doing Atlas programming on the 220 in 1964. In all, 18 Atlas guidance computers were built at a total project cost of $37 million. The computer was very reliable, and no Atlas launch was ever aborted due to computer failure. During the later phases of the Atlas project, Burroughs also developed a computer used in the Polaris submarine-launched missile system.
In 1959 Great Valley began development of a very different sort of computer for the Naval Research Laboratory. The Burroughs designation for it was D825, but the Navy, of course, had a more elaborate name: AN/GYK-3(V). The D825 was a very sophisticated machine which, like UNIVAC's LARC, was designed for the possibility of having multiple processors. The design employed a crossbar switch arrangement where switch modules (distinct from processors or memory modules) contained circuitry to accommodate and queue simultaneous memory requests. Later models of the D825 could have up to four processors, sixteen memory modules, and ten i/o controllers. Since neither of the two LARCs ever had more than one computing unit (CU), the D825 was, according to Philip Enslow, the first true multiprocessor computer. (Philip H. Enslow, Jr., "Multiprocessor Organization - A Survey", Computing Surveys, Vol. 9, March 1977),103-129). The memory used a fairly large word size: 48 bits. For fixed-point binary arithmetic this was broken down into one bit for the sign and 47 bits for the value of the number. Floating point numbers were represented with one bit for the sign of the exponent, 11 bits for the value of the exponent, one bit for the sign of the mantissa, and 35 bits for the value of the mantissa. A 65,536-word drum and tape drives provided auxiliary storage. The first D825 was delivered to the Naval Research Laboratory in August 1962 and was operational by November. Burroughs wrote an operating system called the Automatic Operating and Scheduling Program (AOSP), an assembler, and compilers for ALGOL, FORTRAN, and JOVIAL. JOVIAL was a variation of ALGOL which was widely used on military computers. AOSP allowed a program to have multiple activities ("parallel processes") executing simultaneously on different processors, as well as multiple re-entrant executions of a single program. The second and third D825s were shipped to Systems Development Corporation and MITRE Corporation in the spring of 1963. As a result of a 1963 proposal, another D825 was sold to the Defense Department's Joint Communications Agency at Fort Ritchie, Maryland for message switching. Burroughs wrote a different operating system, Automatic Message Operating System (AMOS), for it to handle large volumes of messages. The Air Force bought 34 D825s for part of the North American air defense system, four more went to Mountain States Telephone Company to handle air defense communications in Colorado, and four were used by the Federal Aviation Agency for a radar control system. The Navy bought at least five more D825s for various projects. In all, it was a very successful system for the company.
While Great Valley was concentrating on military work, the other Burroughs computer development group in California got sidetracked on the 2111 project. The advent of transistors had made their Datatron and 220 models obsolete, and a logical follow-on would have been a transistorized version of the 220. Instead, the engineers lost over a year during 1959 and 1960 working on the 2111, a medium size machine which was to compete with the IBM 650 and the UNIVAC File Computer. This was the wrong way to go, since the 650 and the File were both machines which used the technology of the mid-1950s. Neither one had core memory so the design team of the 2111 decided not to use core memory on it. Their July 1959 proposal said: "Thorough studies early in the program proved that we could not realistically expect to meet the pricing requirements of the 2111 System if the main memory were to be magnetic cores. Of two options, our Engineering and Marketing Divisions were unanimous in their recommendations of the magnetostrictive line technology for the main memory of 4096 words." The pricing requirement was that the computer had to rent for no more than $10,000 per month. Its memory was to be 64 magnetic delay lines, similar in concept to the mercury delay lines of the UNIVAC I (but without the mercury), to store magnetic pulses. The pulses would go in one end of the line and come out at the other end, three milliseconds later. Each line would store 64 37-bit words. Since 4096 words was not a lot of memory, there would be an auxiliary drum with a capacity of 38,400 words. The goal of the 2111 project was to begin production shipments in 1962. Design work continued until March 1960, when the company realized that a computer without core memory would not sell. E.S. McCollister summed it up in a memo to James Bradburn, head of the division: there was "unanimous consensus that the 2111 program should be abandoned" because the "cost/performance characteristics do not fall within acceptable boundaries." (2111 System Proposal, and Memo, E.S. McCollister to J.R. Bradburn, March 14, 1960, both in Technical Engineers' Papers, Burroughs Corporation Records ,CBI 90, Charles Babbage Institute, University of Minnesota). The Pasadena development organization had to regroup and come up with something better. The something better was the Burroughs 5000.
An article on the Burroughs 5000 was published in the May 1992 issue of Unisphere. Copyright 1999, by George Gray