Original URL: http://www.theregister.co.uk/2013/06/26/maurice_wilkes_centenery/

'Flash Gordon' tech: How Sir Maurice Wilkes made practical computers possible

ENIAC - a time before integrated circuits

By Dave Wilby

Posted in Vintage, 26th June 2013 13:44 GMT

Centenary Born this day 100 years ago in Cambridge, Sir Maurice Vincent Wilkes was a pivotal figure in the world of digital computing.

Few would dispute the critical role played by Wilkes in developing practical computing that would ultimately lead to the accessible machines we rely upon today. Certainly for the British computing scene, his contributions were vital.

A brilliant mathematician and physicist, Wilkes was instrumental in establishing the Computer Laboratory at Cambridge University in the 1930s, which he ran following wartime service researching telecommunications and radar systems.

But it was in 1946 he began the work for which he is still best known - the development of the Electronic Delay Storage Automatic Calculator (EDSAC). EDSAC was a general purpose, electronic digital stored-program computer inspired by John von Neumann's (Electronic Discrete Variable Automatic Computer (EDVAC) binary architecture.

Operational in 1949, EDSAC used vacuum tubes for logic, mercury delay lines for memory, punch tape for input and a teleprinter for output.

EDSAC stands out for two reasons.

Importantly, from its inception, EDSAC was developed as a practical computer rather than a blue-sky concept, and this approach typified Wilkes' life work.

Sir Maurice Wilkes photo Computer Laboratory University of Cambridge

Before becoming a Sir: a white-coated Maurice Wilkes with EDSAC. Photo credit: the Computer Laboratory, University of Cambridge

EDSAC was immediately put to work making calculations for Cambridge researchers, and eventually became the basis for the Lyons Electronic Office (LEO). LEO, the world's first business computer, was considered sufficiently robust to run the Lyons catering business to run administrative applications. LEO was soon farmed out to other businesses and became the basis of a national international computing empire, featured in the first installment of The Reg’s Geek’s Guide to Britain.

What also helped make EDSAC unique was its relatively compact build compared to other contemporary electromechanical and electronic circuit digital programmable machines in use, such as the Z3 or the ASCC.

And a major reason for this was EDSAC favored vacuum tubes over switch/relay logic. Vacuum tubes not only meant a smaller machine, because they performed the role previously done by the bigger physical hardware, but they also meant faster, more powerful and more reliable processing combined with the ability for EDSAC to process different types of workloads.

Until EDSAC, machines were limited in the jobs they could perform by their hardware – gears, levers, relays - or the program. But an electron is an electron is an electron, no matter what the program, and the valves worked by controlling the flow of the current in a sealed container.

In a world of integrated circuits and compact design, it’s hard to comprehend how a device that at best resembles a lightbulb and at worst something from a Flash Gordon rocket ship could make a real-life computer run.

Just how was it these glass wonders revolutionised computing and how was it their chapter in that story was so brief?

In this context, a vacuum tube, thermionic valve, or electron tube describes a sealed component controlling electric current through a vacuum. A typical early tube was cylindrical (sort of), constructed from thin glass, and operated as a simple diode - a light-bulb style filament heating up to direct flow in a single direction through a vacuum to an electrode, enabling rectification. Further developments saw more sophisticated three electrode (triode) designs enabling controlled electronic amplification, leading to applications such as switching, so crucial to computational machines.

The big leap forward provided by vacuum tubes was that they meant a shift away from electromechanical computers - which had used levers and gears or relays (Z3) - to electronic computing. They replaced the magnetically operated switching relays carried forward to early computers from telegraph circuits and telephone exchanges.

One of the first computers credited with using vacuum tubes was the Atanasoff-Berry Computer (ABC). It was built in 1937 but was not programmable and was designed primarily to solve linear equations. The Colossus computers, in working order from 1943 and used from early the next year onwards to crack the German military commands during World War II, also used tubes and were programmable. This was also the case with Electronic Numerical Integrator and Computer (ENIAC), which began work for the US Ballistic Research Laboratory in 1948. ENIAC forged the way for a generation of programmable vacuum tube computers from 1949, spearheaded by Wilkes' EDSAC.

Although vacuum tubes ultimately made electronic computing a reality, pioneering engineers were initially concerned about their relative fragility, short life, unreliability, and high failure rate when compared to electromechanical relays.

Fortunately, brilliant Colossus designer and General Post Office engineer Thomas “Tommy” Flowers decided that thousands of vacuum tubes could and should be used reliably in an electronic computer as long as the environment was stable and the circuits were kept on constantly. Flowers had seen vacuum tubes in action at the GPO’s newer phone exchanges.

5,936 glass valves sitting on a wall...

Flowers, who overcame resistance from colleagues, was right and oversaw builds of the first Colossus, which used 1,500 custom vacuum tubes, or valves, as well as the second, which used 2,400. EDSAC used 3,000 tubes and consumed 12 kW of power. The LEO 1 used 5,936 valves with hundreds more in external units, and had a total power consumption of 30kW.

Wilkes had first encountered thermionic valves through his work on radar during WWII and incorporated them into the build of EDSAC.

But, just as electromechanical relays were gradually replaced throughout the 1950s by vacuum tubes, so these too succumbed to technological progress and were themselves usurped as the logic of choice in a new generation of computers.

Although Flowers convincingly proved that vacuum tubes managed sufficient stability to create previously unimaginable electronic computers and Wilkes used the lessons learned to make huge leaps in processing power, vacuum tubes or valves were never entirely practical for commercial usage. In the early 1950s, Lyons technicians had to replace on average s whopping 50 of these valves each week in LEO I – the business sibling of EDSAC.

valves photo by Stefan Riep

Seems logical: vacuum tubes, thermionic valves, electron tubes on display. Photo credit: Stefan Riepl

In the late 1950s and early 1960s, vacuum tubes made way for semiconductor diodes, and transistor circuitry evolved from the first experimental transistors of the late 1940s. “Solid-state” electronics confined charge carriers within solid materials, rather than a vacuum.

As solid-state research progressed, it became clear that it could be used to build devices that were smaller, more efficient, more reliable, cheaper, lasted longer than tubes, and didn't need time to warm up. The UNIVAC, one of the very first solid-state computers, was created in 1958 and used 700 transistors and 3,000 magnetic amplifiers for logic (although it still relied on 20 vacuum tubes to control power).

Transistors and printed circuit boards transformed the computing world in the 1960s but it was Jack Kilby and Robert Noyce's development of the integrated circuit (IC) or microchip (a technology which had been in research since the heyday of vacuum tube computing) that really revolutionised the industry. Chips exponentially increased processing power and made minicomputers a genuine option for mainstream businesses seeking an edge over competitors.

Intel joined the fray in 1971 with 4004, the world's first commercial microprocessor. 4004 was capable of 60,000 instructions per second, compared to the rather remarkable 100,000 instructions made possible by the 18,000 valves in the ENIAC. Modern chips built around largely the same integration can offer processing and storage capabilities that Maurice Wilkes would only have dreamed about. Just last week China's Tianhe-2 supercomputer became the world's most powerful, posting a peak performance figure of 33.86 petaflops.

Vacuum tubes are gone from computers but remain used in a number of relatively niche areas mostly out of preference today - notably in audiophile sound equipment as well as some radio transmitters. They are believed to provide a "better" or "more unique" sound or to be reliable. Because of this, vacuum tubes are still produced, and there is a thriving online in new old stock (NOS) tubes dating back as far as the 1920s.

Don't rule out a return of the tube to computers, though. NOS stocks may become even more valuable one day, such as in the event of a massive attack on the critical computing infrastructure. That’s because vacuum tubes are far less susceptible to damage from nuclear explosions and electromagnetic pulses, unlike today’s transistors and ICs - which would get fried in a second. ®

Bootnote

There's a rich history of vacuum tube and valve computing and plenty of readers with hands-on experience of either working with valves or decommissioning the machines that used them. Please share your experiences and anecdotes in the forum topic on this subject..