I was with Marconi's Wireless Telegraph Co, at Chelmsford, as a graduate apprentice (Cambridge University, Physics) but a family problem forced me to move to Lancashire, resulting in my joining the Electronics Dept of the then English Electric Company (Aircraft Division) in 1956.
Tom Duerden set me to work developing vibration recording equipment (on film, this being pre-magtape) and an analysis system. This was of course analogue. Electronics Dept ran a big analogue computer used for flight simulation and I recall many arguments over the years about which was better - analogue computation or digital. Tom kept visiting Ted Petherick at Farnborough and was very inspired by Petherick's reflected codes for reliable positional digital readout from coded discs. He also kept in close contact with Filton and served on "Dr Russell's Committee" which did crystal ball gazing.
Then Tom said that we were to get a digital computer and that I was going to look after it - he would recruit another graduate to take over my existing project. A 6-month Maintenance course was about to start at Kidsgrove.
DEUCE MAINTENANCE COURSE AT KIDSGROVE
There were half a dozen people on the course, from British Petroleum who were to have a machine in London, English Electric Whetstone and a few existing Kidsgrove staff. A motherly soul known as Mrs P (easier than Poliakoff or whatever her Eastern European name was) looked after domestic arrangements and I lodged at the Alsager Arms.
We had very down to earth classroom instruction (in the style of the Admiralty Handbook of Wireless Telegraphy) from Arthur Bailey. Over several weeks he put over what the computer was trying to do and why, (which was half the battle) then went through logic diagrams step by step to show how it was to be achieved. (I recall that someone objected to the use of logical "Inhibition Gates" as not being in accordance with Standards.) Then we studied the actual circuit diagrams, understanding how valves, resistors and diodes could process pulses to implement the logic.
Considerable time was spent in studying the interwiring, tracing the distribution of signals and controls and could understand the function of every wire. The construction of the room-size frame and the wardrobe-sized power supply was covered. Then we went on to the Hollerith Reader and Punch, with the adjustments to their cams. And the mercury delay lines.... We knew that machine inside out.
In parallel with this we learnt the instruction codes, the fundamentals of
programming and how to input simple tests, either manually from the console or
by punching a few cards. We studied standard test programs, working through
"Functional Test Buzz and Go" line by line of code and realised that
diagnostics could be misleading when the machine was dropping or gaining bits
from the program store! We had some hands-on experience of fault-finding, the
instructor removing valves or replacing them by "Johnnies".
We thought it was all so very sophisticated but looking back it seems so straightforward!
Then we went onto the commissioning floor, to help in building up machines. Chassis were received from the wiremen untested - but there were seldom wiring errors.
Starting with "Unit A" (Oscillator), chassis were progressively inserted into the frame and action checked visually by oscilloscope. Use of the bias box often showed that Select On Test components had to be changed - e.g. - where a nominal 100k resistor had been built as 82k + 18k the critical nature of the circuitry required this to be changed to perhaps 82k + 22k or perhaps 82k + 15k. The replacements were just tacked in and a wireman would fix them to inspection standards overnight.
It was a great milestone when the first two bays of chassis were in, resulting in a complete set of 32 Q-pulses. Thus we continued, getting memory going, Control and being able to make panel lights flash on and off automatically.
Another milestone was being able to run the test program "Functional Buzz and
Go" and then widening the Master Bias margins. Ultimately the magnetic drum
was fitted and Colin Haley from NRL Stafford (its designer) came to set it up
mechanically (ploughed drums are expensive) before the associated circuitry was
I feel that I could just walk back into the job!
All this activity was monitored by Derek Royle, who was a most competent all-rounder. I think that he would officially be DEUCE project leader at Kidsgrove but as well as his managerial ability he did stints on the commissioning floor and some programming.
EXPERIENCE AT OTHER SITES
After 4 months I went to help Frank Thompson to install the BP machine in the City of London. The layout was dreadfully cramped, the Delay Line mushroom having to be placed on a ledge 3-feet high, making adjustment awkward.
I also was sent to Filton to stand in for a week while their engineer was away. Unfortunately their machine got very sick on the Thursday - the monitor display was jumping and test programmes would not read in. What use are diagnostics then? Power supplies and bias lines were in order.
I had to call in the
D.C.M.S.U. (Digital Computer Mobile Service Unit). This comprised two very
shrewd people, Jack Richardson and Jeremy Walker. By the Friday night they had
found the problem; a resistor had gone open circuit in the Clock Pulse
generator, resulting in the pulses being about half a microsecond wide instead
of a third of a microsecond.
I thought that was mean!
I further went to Shorts and Harland at Belfast for a week.
BACK AT WARTON
Meanwhile Tom Duerden had been negotiating for space at Warton for our machine and personnel and arranging for delivery. He never programmed or used the machine himself.
He had assembled a small team of programmers, led by John McDonnell (a Cambridge maths graduate), whose elegance made him stand out from the blunt North-countrymen surrounding him. John Halliday stayed as his No.2 throughout. While waiting for our machine they travelled or sent specially-built metal boxes of punched cards to other installations.
When we had become established, the Department was changed to Electronics and Computing. John McDonnell's team were located upstairs in a different building (unfortunate, because it generated a slight feeling of Them and Us - it's Our machine, not yours). John McDonnell's team performed a general service to the site. Additionally a specialist programmer was resident in each of the main departments - Gordon Sumner in Flight Test, Gordon Pitt in Stress, Maurice Marvin in Aerodynamics. There was also representation in Mechanical Test, Weights, etc.
Tasks comprised typically curve fitting (Runge Kutta) to transducer calibrations and resultant scaling of data to engineering units; framework analysis (matrix algebra using NRL's Scheme B); performance analysis - but all I know is that jobs could take anywhere from five minutes to three hours and comprised cards in, cards out.
Essential to the system, were what were loosely called the 'Punch Girls'. In charge was Josephine (Jo) Lloyd, who was transferred from N.R.L. Stafford, where she had experience of operations. Her No.2 was Barbara Salisbury. Barbara was transferred in from Flight Test, where she had been a Mathematical Assistant -i.e- a computer, doing much calculation by turning the handle of a Brunsviga machine. These two girls led the half dozen who operated the DEUCE for much of its production work.
And here is the good news. In the fullness of time, Jo married Maurice Marvin, and in 1959 Barbara married me!
certificate in fact describes Barbara as a 'computer' !! As we were both being
married, that meant two collections in the department for wedding presents. But
it was against Company Regulations for husband and wife to work in the same
department, so Barbara had to go. In fact, we opted for Barbara not to transfer
to another department but to give up work and become a 1950's housewife.
That meant a collection for her farewell present also - people were nearly bankrupted! We put the three funds together and bought a silver teapot which is still in daily use.
One of the jobs of the "punch girls" comprised using a typewriter-size Hollerith machine to "type" handwritten sheets (program code or data) onto punched cards. A different girl inserted these cards into a larger Hollerith machine and typed the same data again, the machine verifying that exactly the same had in fact been typed, so proving whether the original was correct.
Though this appears inefficient, it saved subsequent wastage of precious DEUCE time. Data cards had to be assembled, together with program cards and perhaps parameter cards, for running on the DEUCE. The computer operator might also have an instruction sheet, detailing keyins.
Some cards were not punched by hand but by semi-automatic machinery procured at the instigation of Tom Duerden. Benson-Lehner machines called OSCAR and BOSCAR, which output a punched card for each reading, were used by Flight Test to measure coordinates on films.
We never added paper tape equipment to DEUCE, though it was used extensively elsewhere. We never expanded to 80-column operation, our tasks being scientific rather than commercial.
Typically I had the machine first thing for maintenance. Then programmers came down to test their recent work (there were no remote monitors).
In those days there was much Single Shotting, whereby the correct sequencing of every instruction was checked and its effect on the registers observed on the monitor. (The scheme of DEUCE instructions was heavily criticised, for example an instruction in minor cycle 28 wanting to add a constant from Store 1 minor cycle 5 to the number in store 13 would have to specify a wait of 7 in the instruction 1-25, which gives plenty of opportunity for a coding error - there was no Assembler language at that time.)
If an error was found, perhaps a further hole would be hand-punched in a card - or a chad would be manoeuvred as a stopper into an incorrect hole and rubbed with a pencil to persuade it not to drop out; the corrected card would then be re-entered and could later (if it was lucky) be copied on the Hollerith reproducer.
Ultimately the program was ready to be tested for giving the correct result from specific cases, before being released for production. Then the girls (sorry to be sexist) would get on the DEUCE to do production jobs.
All DEUCE output was on cards. These were taken to another adjacent room where John MacFarlane operated and maintained the Hollerith tabulator, a huge clanking machine which printed results.
Alternatively the cards could be fed by a girl to a Hollerith reader attached to a Dobbie McInnes plotting table, which I maintained. (This could only plot points. Curves, scales and legends were written subsequently by hand). We had no on-line input or output.
Depending on workloads, there could be a further program testing session in the afternoon, or a night shift organised.
But hold on, I haven't described maintenance. Close to the machine I had my office, spares stores and test equipment. I was variously assisted by Terry Hughes, Fred Davies and Al Beedon, whom I had to train on site. New starters were sometimes sent to me as part of their "induction training".
It was Kidsgrove's firm recommendation that two hours daily "Scheduled Maintenance" should be spent in bias checking. On the machine there were about 1000 sockets at which a voltage offset could be injected. With a program running, the range of correct functioning could be monitored.
By recording the figures, trends could be observed and thus a valve could be replaced before it failed in normal operation. Thus the machine would be reliable.
I thought this somewhat simplistic, as it required that the deterioration rate was steady and much slower than the frequency of testing (nominally once a month). But maybe an hour after being demonstrated to be in good health, a valve would suddenly give up!
Bias checking was tedious, as the program was frequently knocked out and the cards had to be re-loaded. Similarly the magnetic drum would be knocked out of sync. Not all failures were of valves; there were other stock faults.
Dramatically a Bulgin connector would track, shorting a power line to chassis, or a ceramic lead-through capacitor on a power line would go short circuit, making the machine go off with a loud bang. (Other stock faults are described below.)
I came to think that overall productivity was achieved by my daily checks comprising just the brief proving of the machine by running Functional and Magnetic Drum tests at High, Normal and Low master margins, then handing the machine over as a healthy operational machine an hour early. Let it do extra useful work until it called for me! (We always could use extra machine time.) But I was never able to prove my philosophy, which was deprecated by Kidsgrove.
Of course I checked the six bias lines for individual range fairly frequently, and chased up any anomalies. On replacing a valve, I sometimes found that a resistor value had to be changed. (This was despite valves being pre-tested and selected.) Thus, unlike some sites, I seldom had to have the bias box inserted to force production jobs through!
FURTHER STOCK FAULTS
A recurrent cause of failure was mercury getting under the crystal in the mercury delay lines, causing dropping of digits so I frequently monitored the waveform of the emergent signals by oscilloscope.
Another stock problem was random errors in reading from the magnetic drum. This was caused by the reset pot of the head shift mechanism. This potentiometer was wound with extremely fine wire to give the 16 head positions via a pickoff. Material would wear from the pickoff and short circuit adjacent turns, resulting in positional dither and hence misread. Thus I argued that the last thing the machine needed was a dose of test programs - the machine does not learn.
The Magnetic Drum test repeatedly hurls the pickup up and down the potentiometer, worsening the wear. I would have liked the opportunity to have dispensed with this analogue positioner and replaced it by a digital (with mechanical linkages) mechanism, but of course continuous production workload prevented major experimentation.
Thus I spent most of the day in my office, (where I was far from idle but I won't go into that). I would be called to look at the machine typically two or three times a day. Production work would sometimes extend into the night, without an engineer present.
A telephone call maybe at 3 a.m. would bring me in on my motorbike (I lived only 3 miles away). I did not get paid overtime for that but regarded it as part of the job and might take time off in lieu.
Talking of night work, it was not unknown at DEUCE sites for personnel to sleep inside the machine; it was nice and warm. I never heard of anything more extreme.
DEUCE ENGINEERS MEETINGS
An annual meeting of DEUCE engineers was organised by Derek Royle; this was very good for morale and developing camaraderie. I attended those at Kidsgrove, Whetstone, Queens University Belfast and Warton.
Derek would give a presentation on developments - Automatic Instruction Modifier, 64-column. We would have a site tour and discuss our experiences and the various modifications we had made.
I particularly recall Derek Royle's saying that when the first machine was built the decision had to be taken of where to place the DEUCE 32-bit field on an 80-column Hollerith card; if columns 1-32 were used, the card would be unbalanced mechanically and might be prone to misfeed. After all, the Hollerith system had been intended for decimal data, not the heavy punching of binary. (In the extreme case, there were what Derek called "window cards", used in setting the timing of the reader.)
So a more central position for the field was desirable and 21-52 was deemed convenient. How this decision was cursed when 64-column was thought of; there was not adequate contiguous space remaining! Complicated switching had to be used to move the basic (alpha) field down to 17-48, so that the new (beta) field could sensibly be 49-80.
Unofficial modifications were frowned upon, for agreement with drawings and possible conflict with future developments. However, to aid location of bias criticality one site had interrupted each of the six bias lines at the entry to each bay and this got adopted. Also, to speed location of bias criticality, one site had come up with a simple device - just a 1.5v battery mounted on a plug, to offset by +1.5v or -1.5v and these were actually produced by Kidsgrove.
We had both the AIM and the 64 column conversion done by Kidsgrove wiremen, necessarily on site. They also fitted the bay-by-bay bias checking and forced ventilation of the magnetic drum. Names were Ray Morris and Sam ? Their workmanship was superb.
But there were many more minor unofficial modifications. At almost every site the anonymous row of 13 white bulbs showing the current instruction had lenses interchanged with trivial indicators such as Power On, so that being changed to 3 of one colour, 5 of another and 5 of another, the constituent parts of the instruction (NIS, S, D) could more easily be read. The associated keyswitches similarly had their colours changed.
It is useful to see which track of the magnetic drum was being used, particularly during program testing; Kidsgrove procedure was to go round the side of the machine, open the door and view the local indicators. I added repeater indicators on the operator's console and most sites did similarly.
Though initially deprecated by Kidsgrove, I think that this became official and incorporated in later machines. The Hollerith punch on the Deuce itself adds Job number and Card number. At the request of our operators I expanded the Job number to more digits.
Our operators had other tasks to perform while running the machine (quite apart from getting the tea or dozing). They requested that the machine would summon them when it needed them - e.g. - the job needed more cards, or had completed. I devised an electric bell which rang when the "GO" trigger went off except in Reading or Punching. Derek was horrified when he spotted the bell on a visit; "What the hell is that doing here?"
The facility could be switched on just when desired. To avoid drilling a hole I changed a two-position keyswitch on the front panel to a three position, choosing Cont T.C.I. (TIMCI) as the machine never was run with this in the Down position. This subtle choice ricocheted because our operators went to use the Kidsgrove machine while ours was being converted to 64-column and caused consternation by reporting that "The TIMCI key won't go up" - she thought it was a standard facility!
We all know that the DEUCE was made to play musical tones by Software putting the programmable Alarm trigger on and off by 7-24, 6-24 loops, this being fed to a loudspeaker. I took this further and fed a loudspeaker from various points, typically one of the Instruction Staticisers. This resulted in a throbbing noise, similar to Internet on a telephone.
I found this useful. When bias checking, I could tell that the program was still running (and the Functional Buzz and Go test program could not sound the buzzer if it had stopped!) and particularly that the machine was running correctly (if a bit was dropped from the program store, the machine could be operating in a spurious loop). The audio also had some merit operationally. For example, during matrix algebra, we could hear the stage when a matrix was being transposed and so know when the end of the job was approaching or at what stage a failure occurred.
Of the modifications which I devised perhaps the most appreciated was to the CRT monitor. The long delay lines were displayed as a matrix of 32x32 dots. During program testing, programmers could be observed laboriously counting dot positions with the aid of a pencil point! At one site, paper scales were made to put against the glass.
I altered the scan electronically, so that separated subsets could be shown, both vertically and horizontally, allowing ready interpretation of both bit position and minor cycle. The bit position could be displayed in instruction-oriented groups (NIS, Source, Destination, Wait, Joe, Timing, Go) or as data (5,5,5,5,5,5,2) or bytes (8,8,8,8). Horizontal and vertical splitting were controlled by new switches - I dared to drill new holes in the panel! This scheme was copied by several sites and Kidsgrove reluctantly admitted that it had merit.
I observed that we were getting resistor failures in all bays but most
occurred in the bottom chassis. The machine was force-cooled by a large
external blower, from underneath. I fitted deflector plates to stop the cold
blast from striking the bottom chassis, with considerable success.
(Derek Royle again was horrified by such tampering - but at Kidsgrove the machines are run without ventilation with the doors open!)
Test equipment had to be connected to the machine. The official "bias box" was powered externally and the power source had considerable capacitance to ground. The very action of plugging it in caused a transient on the bias line, which could knock the machine out of program, wasting more time. I inserted high value resistors in the connection lead to reduce the effect, a press switch shorting these out when taking a reading.
The oscilloscope very frequently had to be used. Test leads had to be clipped to the circuitry but the turret lugs were closely spaced. Even Kidsgrove personnel often accidentally short circuited to the next lug and with a flash and a bang the machine powered off. Yet we know that the life of valves is shortened by switching them on and off - they are best left undisturbed. So "maintenance" was detracting from reliability! I and most sites therefore added safe accessible test points at frequently used positions -e.g.- output of delay lines, clock pulses. Kidsgrove were not keen, as these could not be built to engineering standards. Also the oscilloscope often had to be referenced not to ground but to the -200 volt power line. To save having to keep clipping to this on a turret lug or component, I bought out a switched scope reference socket.
(In retrospect, we were told to keep one hand in the pocket while clipping on, to avoid electric shock, though I witnessed many. As there was also a +300 volt power rail, very severe shock was possible. I doubt if modern Health and Safety would allow DEUCE.)
Thus my maintenance philosophy was to notice what was causing problems and take action to prevent it in future. Otherwise leave well alone.
I did not do any serious programming, though I devised several useful tests/demonstrations. One which comes to mind was a new Initial Card. The Master Clear key removes digits from the high speed stores but does not affect the magnetic drum. Consequently unused tracks on the drum still carry program/data from a previous job.
If Head Position was not applied correctly, this would be accessed into the current job, causing chaos. My scheme wrote zeroes on every track of the drum and left the Head Positions set to 0, also performing all the actions of the standard Initial Card. This was achieved with a single card, by exploiting the structure of the DEUCE instruction set. Derek Royle said that he found it intriguing but that as it assumed that unused bits in the instruction word would have no action (c.f. "Joe") it might give compatibility problems if DEUCE became further developed, so it was not adopted.
Visitors were always impressed by the imposing power supply cabinet, with its six large voltmeters and six ammeters. They were intrigued by the switches marked Cancel On and Cancel Off (associated with a time switch). What happens if Cancel Off is set to On? And it has to be decided whether the switch is up for On or down for On, the label being above it! I relabelled the switches logically with the legends Automatic On and Automatic Off, though we did not use the automatic facility.
THE FINAL PHASE
Workload increased, so we got a second machine. Our room space forced the machines to look at each other (one operator between them), so we could not take a nice photo as at Farnborough.
The second machine was 64-column and our first machine was then upgraded to match. We treated them similarly and found no difference in their reliability. It would have been nice to be able to use them to prove differing maintenance philosophies but there was much serious work to be done.
However, Tom Duerden deemed that I must move on and tried to "de-DEUCE" me by giving me extra little projects. I was supposed to change to a desk back in Electronics Department mid-morning. I hated this because I could be busy enough with my two "babies". Now having computer in my blood, I would have applied to transfer to Kidsgrove to work on KDF9 if my family tie have not still been paramount.
THE END OF AN ERA
Then in about 1961 Tom Duerden told me that magnetic tape would be used for airborne data recording on our new TSR2 aircraft. I was commissioned to design and oversee the build of a ground station to suit. This was a huge task, so I completely left DEUCE, leaving Al Beedon in charge.
My Flight Test ground station occupied 20 six-foot cabinets and was wholly analogue. The engineer at Shorts and Harlands was playing with the idea of feeding flight data on line to his DEUCE in the style of USA aircraft and space programs.
At Warton we progressed to on-line computer-based processing only for the Jaguar and subsequent aircraft; and that was using a computer larger and faster than DEUCE.
Back in the Computer Department it was not long before it was thought that DEUCE itself was outmoded. There was talk of getting a KDF9 but at Board level the decision was taken to follow the swim and change to IBM. A huge machine - all of 32k - was then obtained.
It was a sad day when the scrap metal men rolled our machines out and took them away on a lorry.
Al Beedon and Maurice and Jo Marvin, moved to Kidsgrove, to what had then become ICL.
Tom Duerden decided that he would have more scope in USA and went - I think - to Boeing.
In conclusion, I am grateful to DEUCE for giving me the happiest days in my career and a lovely wife too.
© Steven Allcock - 8 November 2004