If you think maths is a chore today then spare a thought for the number-crunchers of the 1940s.
Every sum took a physical as well as a mental toll, with each totting-up demanding a crank of a mechanical calculator.
Unfortunately the mathematicians stuck with such cumbersome technology were also tasked with working out how to build the UK’s first nuclear reactor. This was an engineering task requiring long chains of calculations and frequent checks, such that it was not unusual for that handle to be cranked tens of thousands of times.
Beyond being a monotonous and arm-numbing endeavour, this manual approach was inherently unreliable. Every calculation was prone to error, which didn’t sit well with the low tolerance for mistakes when harnessing energy from a sustained nuclear reaction.
The workers at the Atomic Energy Research Establishment (AERE) in Harwell, England needed a better way, a way to automate the grunt work.
“It’s the scientists themselves who problem solve and decide to build this giant computer, what, for them, would have been a large programmable calculator,” said Chris Monk, learning co-ordinator at The National Museum of Computing (TNMOC).
At a time when there were only a handful of computers in the country, the team were fortunate to have such a machine nearby. The Electronic Delay Storage Automatic Calculator or EDSAC had recently been completed at Cambridge, just 60 miles from Harwell. Its creator Maurice Wilkes held regular meetings to discuss ideas about the new field of electronic computing and the Harwell team were regular attendees, squeezing themselves into an open-top sports car to make the journey – often in bitterly cold weather.
Armed with an understanding of the principles and components needed to build a computer, the team began construction of the Harwell Computer in 1950. It was completed in little over a year and without drama, save for one of heads of department at the AERE being jailed for supplying the Soviets with models of the first hydrogen bombs.
The researchers now had a way to generate tables of values for engineers building the nuclear reactors without days of manual work.
But the advantage of the Harwell Computer wasn’t speed, because while a modern computer can perform 100,000 mathematical calculations in less than one second, the Harwell machine would typically take 17 days.
Such was its low speed, a practiced operator of a hand calculator would be able to keep up with the Harwell Computer for a time. But while the person would start to flag and make mistakes after half an hour of cranking a calculator the Harwell Computer would keep going. In the event of a problem, it was built to try the same calculation three times before stopping and sounding an alarm.
“You could go off on holiday and come back and be confident it would still be going,” Monk said, stressing the importance of the machine’s unflagging precision.
“Whereas the Colossus [computer] was built for speed in the middle of World War Two, this would have been more built for accuracy. If they made errors, these errors would be transferred into the construction of atomic reactors.”
While the Colossus is famous today for being the computer that cracked the Lorenz codes used by Hiter and his high command, the Harwell Computer had a crucial advantage – it could load instructions into memory.
That memory was infinitesimal by today’s standards – about one third of a KB – and could only hold about 90 decimal values.
But having a memory allowed the machine to move closer to being a programmable machine with branching logic, similar to modern computers.
“This was significant because the ability to store a program in memory meant could you change it on the fly. You could make decisions and jump to different places based on those decisions,” said Monk.
While the Colossus computer may have been famous for the heat radiating from its valves, the Harwell Computer was a cooler customer.
The bulk of the panels making up the machine were covered in Dekatrons, gas-tubes used for counting to 10, which didn’t heat up when used. Dekatrons were a natural choice for the scientists, according to Monk, as they were already common equipment at the research centre for measuring radiation detected by Geiger counters.
For those interested in how computers work however, Dekatrons can be quite revealing. Watch the rebuilt Harwell Computer in action, at The National Museum of Computing (TNMOC) in Buckinghamshire, England, and you glimpse the fundamental processes that underpin today’s apps.
Every time the Harwell machine stores or fetches a number from memory to be executed, the machine’s 90 rows of Dekatrons click into life and pulse orange as the machine cycles through values of 0 to 9.
Students regularly visit the national museum and Monk uses the rebuilt Harwell machine to demonstrate the fetch-execute cycle that occurs every time a computer retrieves an instruction from memory.
“Because it’s so slowed down, you can show some of the principles of how a modern computer works,” he said.
The other main component of the machine are telephone relays, with 468 type 3000 relays directing the processing and storing of data. This reliance on electromechanical components slowed the Harwell computer down. However the machine was prized for its reliability relative to other valve-based computers at the time, much of which is down to its cool running temperature. Because the Harwell machine’s parts are subject to less heat stress they fail less frequently and some of the Dekatrons on the rebuilt machine at TNMOC hail from the 60-year-old original computer.
While it may date back generations, the Harwell also had some pretty modern tricks up its sleeve. Like other computers from the time, the machine’s instructions and data were loaded by reading binary from punched tape. When a set of instructions needed to be carried out repeatedly, the machine could be set up to run this group of commands on demand – similar to how instructions can be bundled together into a function using a modern programming language.
Doing so required dedicating a tape reader to running this collection of instructions, which would be punched on paper tape stuck together to form a loop. Unlike modern programming these subroutines could only run for so long before the tape would wear out and need replacing.
As new computers came online the Harwell Computer would eventually outlive its usefulness and in 1957 was passed on to a college in Wolverhamption.
The college won the computer in a nationwide contest to give away what the press then called an “electronic brain”.
The machine was kept in service as the WITCH (Wolverhampton Instrument for Teaching Computing from Harwell) until 1973.
But the legacy of the machine’s work would last into the 21st century. Among the computer’s many tasks, it handled more than 34 simultaneous equations needed to design the world’s first atomic reactor for generating electricity. That reactor, Calder Hall in Cumbria, was only decommissioned in 2003, after nearly 50 years of use.
The Harwell Computer is proof that slow and steady really can win the race.