programming. On a conventional machine instructions were stored in consecutive locations, so that latency effects dramatically slowed up the machine. In the ACE design, however, each instruction specified the location of its successor. In this way instructions (and numbers) could be placed optimally in the memory so that they emerged from a delay line just as they were needed. The optimum coding idea showed a rather direct connection to the “instruction tables” of the Turing Machine, in which each order specified the next to be obeyed. This simple idea alone would make the ACE about three times faster than other delay-line based computers. Optimum coding, it should be noted, was much more difficult than conventional programming, but for a decade it was the way to get maximum performance from delay-line or drum-based machines.

But there was more. In the ACE design Turing avoided having a single central register (known as the accumulator in EDVAC-type machines) in which all computation took place. Instead, he had separate registers for addition and subtraction, multiplication, logical operations, shifting, and so on. To a degree Turing was anticipating what would later be called the von Neumann bottleneck.

aCe Construction

bendix g- 15 broChure iMage Courtesy oF the CoMPuter history MuseuM

Turing proposed that the ACE would contain 200 delay lines in the main memory and estimated the constructional cost at £ 11,200 (about $45,000). Turing was a better mathematician than an engineer and his estimates were wildly unrealistic. Building the ACE got off to a shaky start and not much happened for sev-

Besides the DeuCe,
the Pilot aCe
spun off several other
worthy derivatives.

eral months. Then in 1947 an American, Harry Husky, who had been a member of the Moore School computer group, began a sabbatical year at the NPL. He proposed building a small prototype of the ACE, which he called the Test Assembly. Still, little progress was made and Turing left the NPL in September 1947 for a sabbatical year at King’s College, Cambridge University. He never returned to the NPL. At last in early 1949, work really got going on another small version known as the Pilot ACE and the development logjam was broken by Donald Davies, a new recruit to the NPL. Davies was an outstanding scientist and administrator. He later became head of computer science at the NPL and was one of the inventors of packet switching technology.

The Pilot ACE sprang into life in May 1950. It had just 10 delay lines, but when augmented with a drum store it was a highly capable computing machine. Benchmarks showed that it was five to 10 times faster than contemporary machines. Moreover, the Pilot ACE had just 800 tubes compared with 3,000 in the Cambridge EDSAC and 3,500 in the Manchester Mark I.

aCe Derivatives

The English Electric Company, a manufacturer of airplanes and electrical equipment and a contractor to the NPL, decided make a copy of the Pilot ACE; unsurprisingly, the machine was named the DEUCE. This set the ball rolling—English Electric’s competitors decided they would like a machine too and the company found itself in the computer business. Sold from 1955, the DEUCE was especially popular with engineering companies that had heavy computational needs. Sales to aerospace companies were given a boost following the Comet air disaster of 1954, which had resulted in the statutory requirement for airplane manufacturers to conduct so-called flutter calculations. The DEUCE came with a superb library of matrix software developed by Jim Wilkinson, originally Turing’s assistant but later a distinguished numerical analyst and winner of the 1970 ACM A.M. Turing Award.

Besides the DEUCE, the Pilot ACE spun off several other worthy deriva-

a brochure for the Bendix G- 15 computer, introduced in 1956.

 

tives. The most important was the Bendix G- 15 introduced in 1956. The machine was designed for the West Coast’s Bendix Aviation Corporation by Harry Husky, who had returned to the U.S. and joined the faculty of the University of California, Berkeley. Optimum coding gave the machine an edge over rival machines, and the G- 15 became a workhorse for the American engineering industry. About 400 were eventually sold.

Conclusion

As to the ACE itself, the full-scale, 200-delay-line machine was finally completed in 1958. Sadly, by the time it was built the heyday of delay-line storage was over, and new computers were using random access core memory. Even the NPL admitted the machine was a technological dinosaur. Turing died in 1954, so he never lived to see either the triumph of the DEUCE or the whimper of the full-scale ACE. English Electric went on to become a cornerstone of the early British computer industry. After the DEUCE went out of production in 1960 it was succeeded by the KDF9, an innovative machine with a stack-based architecture—it showed the company had lost none of the design verve it had inherited from Turing.

 

Martin Campbell-Kelly ( M.Campbell-Kelly@warwick. ac.uk) is a professor in the department of Computer science at the university of Warwick, where he specializes in the history of computing.