|LEO - Lyons Electronic Office|
|Friday, 19 March 2010|
Page 2 of 2
After initial training and a time at Cambridge Pinkerton returned to Cadby Hall, Lyons headquarters and factory complex in Hammersmith Road, London, and started work with Lenaerts. They ordered an oscilloscope, some benching and tools and set to work. They experimented with transmitting pulses down wires and built a mercury delay line. At the time this was the only reliable form of fast storage. A serial bit stream was converted into acoustic pulses which travelled down the mercury line only to be converted back to a bit stream and re-injected.
A lot of the hardware was bought from electronic surplus shops that occupied most of Lisle Street in London until the late 1960s. After a while they managed to get pulses circulating and implemented a half adder. Much of the design went on like this - building module, proving that they work and then putting them together to create something larger.
The overall design of the machine resembled the EDSAC built in Cambridge - they used the same type of delay line and basic logic, but it was bigger.
The machine, now called LEO - standing for Lyons Electronic Office - had twice the memory and more importantly tackled the problem of I/O.
For a scientific machine such as EDSAC I/O represented a modest problem. A couple of paper tape readers and teleprinters was all that was needed to input programs and output the inevitable table of numbers that was the result. A business machine like LEO needed much more. It needed multiple I/O channels that could run in parallel - this indeed was one of the innovations incorporated into LEO.
Another difference that might seem unimportant from today's standpoint was the need to do arithmetic on non-decimal currency. Scientific computers could work in binary or binary coded decimal and conversion was easy. Today we would solve the problem of working with pounds shillings and pence via software but Pinkerton decided to use a hardware conversion unit built by STL, Standard Telephones Laboratories, a subsidiary of the Standard Telephones Company. This was based on neon gas trigger tubes which worked well in theory but not in practice. The unit was unreliable and eventually removed.
STL also attempted to build a "crush" pack tape deck. Instead of winding the tape onto reels it was formed into a continuous loop and allowed to crush into a storage bin. This allowed the tape to be moved passed the read/write heads very fast and without the need to rotate heavy reels. Unfortunately neither invention worked reliably and they were removed before LEO I was engaged in serious work.
A crush pack tape deck - which didn't work. The same technique was used to create the Sinclair Spectrum Microdrive - and some would say it didn't work then!
Much of the basic engineering design of LEO was borrowed from other machines. They used a system of marginal testing to weed out valves that were about to fail by raising voltage levels so that they did fail - this was borrowed from the Whirlwind project. They used short delay lines as I/O buffers, a rare but not new approach to the problem of handling I/O. Even the exact design of the delay lines was borrowed from the EDSAC.
LEO was soundly built - large racks with spare space and an air cooling system.
But LEO was where many of the basic approaches to business computing were developed first. For example, its engineers had to work out how to get the data in accurately. They eventually developed a system of second keying. The data was keyed to a paper tape and then it was keyed in again while running the tape through a comparitor to detect differences. Much later they even developed an optical mark reader, called Lector, which directly transferred data from specially printed forms into the machine. Because of the unreliability of the machine it was also necessary to occasionally produce a "dump" of the work done so far so that if it did crash the run could be restarted from the last dump rather than from the very beginning. In software design they also very quickly developed modular programming and used automatic relocation.
The scientific computers of the time had little need for this sort of approach because their programs were short and complicated. They used subroutines to implement new functions such as SIN or COS and then used these in fairly short main programs as if they were new instructions. The need to build large programs by picking modules from a library is something that only becomes essential when you are writing the sort of data handling procedures implemented on LEO.
A business first
LEO was a success. It worked and it ran the world's first commercial data processing operation. In 1951 it processed the Lyons bakery valuation data and went on to handle payroll, ordering, stock control, tea blending and so on. Soon after the huge valve serial computer was completed they started work on LEO II and then a transistorised LEO III.
Lyons set up a subsidiary with Pinkerton as director to sell LEOs to the world. They had some modest success but their customers were suspicious of a company that had no background in electronics. In the end they were merged in 1964, with the English Electric Company.
In 1965 LEO I was switched off after 14 years of service at Lyons headquarters. In 1968 English Electric merged with ICT to form ICL and the age of the small UK computer manufacturer was over. Lyons later went on to to buy IBM equipment in order to be able to run their data processing.
Photographs of all the successive Leo machines, and even samples of the noises produced by two of them, can be found at the Leo Computers Society website.
See the side panel for details books about Leo.
They include LEO the First Business Computer which gives an account of the early days of Lyons, the reasons for building LEO and a great many details of the technology. It can't be described as a gripping read because it is an attempt to document the history rather than create a documentary style novel.
|Last Updated ( Friday, 11 November 2011 )|