The UNIVAC 1102, 1103, and 1104

From Folklore
Jump to: navigation, search

Unisys History Newsletter.
Volume 6, Number 1
January, 2002
by George Gray

Remington Rand's Acquisitions

Remington Rand became the leading (and for a brief time, the only) computer vendor in the world through its acquisitions of the Eckert-Mauchly Computer Corporation (EMCC) in February 1950 and Engineering Research Associates (ERA) in December 1951. In addition, the company had its own tabulating machine research and development effort in Norwalk, Connecticut. These three organizations continued to function with a high degree of autonomy, indeed rivalry, for several years. At some point in 1953 or 1954, the company decided to use the UNIVAC name for the products of all three groups, so the original UNIVAC computer was re-designated the UNIVAC I and ERA's 1101 became the UNIVAC 1101. At the time of its acquisition by Remington Rand, ERA was working on a successor to the 1101, which had been called the ERA 1103, and it was later referred to as the UNIVAC Scientific Computer or the UNIVAC 1103. Norwalk's 409 calculator (actually a small plugboard computer for card processing) was sold in two versions called the UNIVAC 60 and the UNIVAC 120.

Having multiple groups gave Remington Rand a split personality. The former EMCC people in Philadelphia were oriented toward business data processing, so the UNIVAC I used decimal arithmetic and emphasized high-volume input/output via magnetic tape. Their leader, J. Presper Eckert, tended to push technology to, and even beyond, its limit, designing machines which were innovative, but not always conservative in terms of engineering practice. Willis Drake (from ERA) described the UNIVAC I as a "rat's nests of wires", and it sometimes required a long time to get one installed and working properly at a customer site. It should be mentioned, however, that Eckert's designs did provide for extensive duplicate circuitry, so that all calculations were double-checked to ensure accuracy. The computers designed by ERA in St. Paul for cryptologic purposes were suited for scientific computation, since they used binary arithmetic, but their input/output capabilities were limited to paper tape and low-speed typewriter. Constrained by rigid Navy specifications, their computers were engineered very conservatively and could be installed and made operational very quickly.

The existence of these two different groups together in one company could have made possible a great cross-fertilization of ideas, and this did happen to a limited extent. For example, Philadelphia's magnetic tape drives were soon added to the St. Paul computers, and St. Paul's magnetic storage drum technology was used in some of Philadelphia's computers. However, it would have taken leadership to really make the two divisions work together, and this was totally lacking. The week after the agreement to buy ERA, Leslie Groves, a retired Army general who was Remington Rand's chief of research and development, went to St. Paul to see just what had been purchased. His blunt manner alienated the entire ERA engineering staff. Norris and Parker complained to James Rand, and as a result ERA continued to operate as a wholly owned subsidiary company until December 1952. Even after it became a division of Remington Rand, it was not placed under General Groves. This meant that there was no coordination between ERA in St. Paul and EMCC in Philadelphia at any level lower than James Rand himself. Significantly, Rand never did visit St. Paul to see the ERA facility. Not until 1955 after Remington Rand merged with Sperry Gyroscope to form Sperry Rand, was one man, William Norris from St. Paul, put in charge of the two computer development divisions. Until then, both reported directly to James Rand, who was 66 years old in 1952. Though he was an astute businessman in general terms, Rand had no real understanding of computers.

The Atlas II

Even before the completion of the Atlas (1101), the Navy CSAW asked ERA in St. Paul to start on the design of a more powerful machine that would use both electrostatic and drum memory. This project was called Task 29, and the computer was designated the Atlas II. Work got underway in 1950, and to supplement its staff ERA hired many of the 1950 and 1951 electrical engineering graduates from the University of Minnesota. One of them was Seymour Cray, who in later years became famous for his work on supercomputers at Control Data and his own companies. The Atlas II project was led by Jack Hill and Frank Mullaney from the Atlas I team, and again Arnold Cohen did much of the logical design. The Atlas II had a 36-bit word, with electrostatic high-speed memory and a magnetic drum for medium-speed memory. The electrostatic memory consisted of 5-inch diameter cathode ray tubes known as Williams tubes, for their English inventor. An electron beam was scanned across the phosphor-coated screen of the tube. If a large current were directed to a spot, it would leave a residual charge that could be sensed on the next scan. These charges had to be continually read and refreshed. Each tube had a 32 by 32 matrix, amounting to 1024 bit positions. The Atlas II had 36 tubes, giving a memory of 1024 words. The electrostatic memory unit was a separate cabinet standing about eight feet high with a six-by-six matrix of Williams tubes, looking like a cluster of ship portholes. The drum provided 16,384 words of memory, with a maximum access time of 17 milliseconds, and was directly addressable as an extension of the main memory: addresses 0 through 01777 were in electrostatic memory, while 040000 through 077777 were on the drum. The Atlas II was delivered to NSA in September 1953.

The 1102

While the Atlas II project was underway, ERA began work on a computer (eventually called the 1102) for the Air Force's Arnold Engineering Development Center in Tullahoma, Tennessee in response to a request for proposal issued in 1950. The Air Force needed three computers to do data reduction for two wind tunnels and an engine test facility. The contract was awarded in October 1952. The 1102s were a variation of the 1101, using its 24-bit word and a smaller (8192-word) drum memory. The circuit packaging and chassis were adapted from those being developed for the Atlas II. Each 1102 had 2700 vacuum tubes, weighed 14,000 pounds, and occupied 122 square feet of floor area. They were connected to data channels coming from the wind tunnels and the engine facility. There were five typewriters for printed output, five paper tape punches, and four pen plotters to produce graphs. The three 1102s plus some peripheral equipment were delivered between July 1954 and March 1956 at a total price of $1,400,000. Software for the computers was developed entirely by personnel at the Arnold Engineering Development center. All programs were done in 1102 machine language, and no assemblers or compilers were ever developed.

The 1103

In the summer of 1952, ERA asked the Armed Forces Security Agency (the predecessor of NSA) for approval to sell the Atlas II commercially. Permission was given, although ERA was required to take out several instructions for the commercial version, which was designated the 1103. Up until this point, Remington Rand executives had not known of the existence of the Atlas II, because they did not have the requisite security clearances. William Norris and several other ERA managers journeyed to the corporate headquarters at Rockledge, an old mansion outside Norwalk, Connecticut, to try to convince Remington Rand's top management that it would be worthwhile to sell the 1103. They wanted to show that it had capabilities which were different from those of Philadelphia's UNIVAC I. Their presentation compared the 1103 with IBM's first computer, the 701, which had been announced in May of that year. James Rand came in late in the presentation and apparently was taken with the idea that the 1103 was a worthy rival to the 701. During several decades of competition with IBM in the punched card tabulating market, Remington Rand had acquired the reputation of being technologically inferior to IBM, and Rand was always looking for ways to even the score. Rand gave ERA permission to build two 1103s and buy the parts for two more. Remington Rand announced the 1103 in February 1953. The first one was sold to the Air Force for use at Eglin Air Force Base (Florida) in its ballistic missile program. ERA sent Erwin Tomash to the 1953 Spring Joint Computer Conference in Los Angeles to publicize the 1103. He gave a presentation in the Remington Rand sales office, and the room was packed: aerospace firms were very interested in the 1103. As part of the Remington Rand reorganization that moved John Parker to New York to be head of all computer sales, Tomash transferred to Los Angeles to handle west coast sales. Very quickly he had orders for 1103s from Convair, Boeing, Lockheed, and the Army's White Sands Missile Range. Unfortunately, the St. Paul factory had trouble making the transition from custom building computers for the NSA to general assembly-line production, and deliveries of the 1103 were late.

The 1103A

These first five 1103s all used electrostatic high-speed memory. Electrostatic memory was a notoriously unreliable technology, and it was regarded as a stupendous achievement when the 1103 at Convair was able to run 50 consecutive hours one weekend to do an important program. A group at ERA began exploratory work on core memory soon after MIT's Jay Forrester published a paper about it in January 1951, and in July 1952 the NSA asked ERA to develop a core memory version of the Atlas II (1103). Since ERA and MIT were both working on defense contracts, ERA had access to MIT's reports and held technical discussions with Forrester's staff. The core memory 1103, delivered in November 1954, had 36 planes of 32 x 32 cores (giving 1024 words) with a cycle time of eight microseconds. The memory occupied two boxes, each being 8 x 6 x 2 feet in size. The design was not suitable for mass production, and was reworked starting with the tenth 1103, where the memory was expanded to 4096 words, and the computer was now called the 1103A. The 1103A cores had a 0.08-inch outside diameter and a 0.05-inch inside diameter. The first 1103A was installed in September 1956 at Lockheed Aircraft at Palo Alto, California and the second at Boeing in November. The 1103A was further modified to provide nine optional floating-point arithmetic instructions in addition to the 41 standard instructions, and this version was called the 1103AF. The 1103 instructions used a two-address format: a six-bit operation code followed by two 15-bit operand address fields. The 1103A had the capability to use one or two additional core memory cabinets of 4096 words each, giving a maximum of 12,288 words of core memory. Up to ten UNISERVO tape units (the type used on the UNIVAC I) could be attached to an 1103A. There were also punched card and paper tape devices. Since the planning of the Atlas II had started prior to Remington Rand's acquisition of ERA, the ERA engineers had arranged to use card readers and punches supplied by Bull. Field engineers found them to be a major headache. Another new feature on the 1103A was a program interrupt capability, which had been developed on an 1103 at the NACA (forerunner agency to NASA) Lewis Research Center in Cleveland. It permitted the 1103A to interrupt the processing of one program to handle another. The 1103A was a large machine, using 3900 vacuum tubes (plus 470 in each memory cabinet) and taking up 58 by 30 feet of floor space. A typical 4096-word system cost $950,000, and an additional memory cabinet cost $200,000.

Lockheed, located in Sunnyvale California, was one of the first group of customers and after 1958 had two 1103AF computers, each with 8192 words of core memory. The Operations Research Office (ORO) of Johns Hopkins University, located in Bethesda, Maryland, which did contract work for the Defense Department had been renting time on UNIVAC Is at various government agencies. It chose an 1103 over an IBM 701 because it could be delivered sooner (June 1955) and was less expensive. ORO rented the 1103 on a single-shift, 40-hour per week basis, with any downtime on one day being made up by extending time on another. They used it to do war game simulations, employing an attached CRT screen made by Stromberg Carlson, and to run other batch work. In 1957, ORO replaced the 1103 with an 1103A having 4096 words of core memory and added six Uniservo tape drives to the system, for a total monthly rental of $24,838. Since the war game simulations frequently had long intervals between inputs (while the war gamers figured out what to do next), ORO made extensive use of the interrupt feature to allow batch work to run in between war game activities.

The University of Minnesota acquired an 1103 as a result of a creative partnership with Remington Rand. Since the machines were being built in St. Paul, the company offered the university 400 hours of free time on computers in the factory during the 1955-56 academic year. This was a strategic move, because it derailed a plan for the university to get an IBM 650. Marvin Stein, who had worked on the 1103 at Convair, moved to Minnesota to become an assistant professor of mathematics and oversee the use of the computer time. Based on his experience, "the 1103 was clearly a superior machine [to the IBM 701] in every respect." The gift of computer time was renewed for the 1956-57 year. After that, the company offered to sell an 1103 to the university for $250,000, roughly one-fourth of the list price. The university accepted, after turning down a proposal to jointly develop a computer with the newly formed Control Data Corporation. The Johns Hopkins Applied Physics Laboratory in Silver Spring, Maryland, Southern Methodist University, Wright-Patterson Air Force Base (Ohio), and Holloman Air Force Base (New Mexico) also used 1103s.

The 1104

A team led by Frank Mullaney and Noel Stone developed the 1104 as a 30-bit variation of the 1103 for Westinghouse Electric Corporation, which was a subcontractor to Boeing on the BOMARC missile program. Boeing was building the BOMARC air defense missile for the Air Force, and it needed a ground-based computer to serve as part of the control system for the missile. The 1104 had specialized input scanning equipment, analog-to-digital conversion capabilities, and special output equipment to feed directly into data communications transmitters. One 1104 was delivered to Patrick Air Force Base in Florida. Once the 1104 project was well underway, the Westinghouse people learned that there was such a thing as the 1103 and asked if an 1103 could have been used for the BOMARC. St. Paul answered that it could have, but it had just followed Westinghouse's specifications. The Defense Department chose the Army's Nike missile to be the primary air defense missile, but the Air Force did deploy the BOMARC to six sites during the mid-1960s. By that time a more modern computer developed by St. Paul had replaced the 1104.

Copyright 2002, by George Gray