Notes from a talk by Stuart Fyfe at The Science Museum Thurs 20 Apr 2000
for The Computer Conservation Society.
The inserted thumbnails are of the slides used in the talk, click to see the full image.
Galdor - Keeping old warhorses in action
I was sharing a flat, over a shop, with a couple of other students in Surbiton for Kingston Polytechnic in the early 70's.
We did a load of weird student stuff - like the electric toilet flush that signalled to the world when it was being used, and could be operated remotely for giving surprises; and the meccano rover that ran up and down the pavement of the main road. Later, there was the model railway that ran all round the flat. We thought it would be fun to buy a scrap computer to play with, and let schools, researchers and ordinary people come and use it for their own projects. At that time, "The Amateur Computer Club" was among the few national groups pursuing this new, strange, hobby. They published hardware designs, and I had already made an adding-machine using old Strowger telephone parts, but wanted something better, to really use. We wanted to link it by telephone to an amateur astronomical observatory, with a big telescope near Croydon. We looked at Elliot 802 with mercury delay line, and a PDP whatsit, and got to know the surplus electronics trade, and the world of 5 track paper tape, but nothing was quite right, not a proper machine.
( Web masters Note )
If I had known then, just what I was getting into, would I have still done it all?
( Oh YES! )
It was a surprise when I had a 'phone call and visit from Rodney Brown, an ICL engineer, who said London University would be scrapping a machine to replace it with a 1904A. This was in the administration building, Senate House. Rodney said it was a bit big, but has everything necessary, and there's scope for adding attachments of our own.
Just the job.
1301 in Senate House, general view with 1904 in background
Big, wow! We visited Senate House and saw "Flossie" in action. Apparently, it was serial number 6, the first one out of the factory in 1962, and boasted the first proper commercial printer.
front view of printer
I think my GCE pass-slips were printed by it, as perhaps, were some of yours. It was, in fact, the programmable tabulator that Babbage wanted.
Apparently, we were only competing with scrap dealers for the price of the metal, and the operations manager, Mr Hutt, was pleased to see it going for educational use at the price of £200.
making coffee on the floor of the computer room
We spent a few days camping in his basement while we dismantled it. Engineering students seem to enjoy this sort of thing, very odd. We'd studied the manuals and knew more or less what we were doing by then, with help from friendly ICL engineers. The units come apart very easily, after cutting the wire wraps between the bays and unbolting the frames. There was no point unwrapping the links as the wires would still need replacing, and cutting was quicker.
disconnecting processor bus bars
And it's all on wheels, which is a very civilised arrangement. There's a lot of ironmongery for the kickplates and covers, so we took photographs to help the process of reassembly. We didn't know how long it might be before it could be turned on again. In the event, it was a couple of years. Well, in 1970 we had nowhere to put such a thing of course, and had to find a temporary store while we looked for permanent premises. There were a dozen friends helping. We strapped the units into a big lorry and
took it to a double garage in Sutton, where my colleague Trevor Jarrett was living.
The last I heard of him, he was selling computers for ICL in Australia. It was a tight squeeze, and there was a small heater to discourage condensation. That house and garage has been demolished now.
It was just across the road from where Harry Seccombe lived. I won't describe the tortuous search for premises, but we had noticed that estates of war-time pre-fab houses were being dismantled. You could get one of these for £200, delivered in sections. It suddenly dawned on us that there was just enough room for one of these in the back garden of our student flat, if we : Bought the lease, then pulled down two derelict workshops, arrange planning consent, and barrow a couple of lorry loads of concrete down the narrow passage that was the only street access.
The best part, was the party for all our helpers and friends in the empty room. The council had made us add an internal brick wall for soundproofing, and there was a lot of fuss about obtaining three-phase mains.
street front of the Galdor Centre
By now we had formed the "Galdor Company brackets Electronics Limited", trading as Galdor Computing. My colleagues Andrew Keene, Derek Odell, and myself were studying electrical and electronic engineering, which included light current and heavy current studies, which seemed quite appropriate.
Others came to our flat or lived nearby, notably Dirk Koopman, Don Hedges, Chris Hewitt, Adrian Barnes; and later Peter Singleton and John Skeene.
The name "Galdor" is a character in a book that I was reading:
Lord of the Rings by JRR Tolkein ... which was required reading for students. A funny thing, we asked for an ordinary telephone line, and without prompting, the number that we were given was 399 1300. I took this as a sign that we were doing something right.
Moving the machine to Surbiton in another tail-lift lorry was easy enough. Except for one unit that was too wide for the passage way.
console on the pavement, with hacksaw
So we got out a hacksaw and attacked some runner extensions.
These runners were temporally attached to the main console - to stop it falling over on its face. That was done outside, on the pavement, in the main shopping street, at midnight.
The most time consuming part of re-installation was, of course, re-connecting thousands of wrapped wire links between the bays.
We made a jig, and manufactured pre-formed links in 3 sizes with insulated wire, and, instead of wrapped joints, we soldered them. And they were colour coded - all green.
(That’s supposed to be a joke by the way).
Soldering made subsequent moving & re-assembly easier for the next owner. We discovered that some corners are best connected by a left handed person, and others by right handed. There's no risk of making a wrong connection, it's all 1 to 1 links. The design is ideal. It's a lovely machine to work on.
rear of printer, plus Andy & deck 7
Instead of the standard in-line arrangement of printer and console, we angled the printer backwards about 45 degrees to make it fit in the building.
The triangle that this opened up allowed a convenient extension of the console work surface.
The first power on was not a success. No smoke, but the clock waveform was all over the place. This derives, more or less, from simply amplifying a drum track. We wanted to load some test programs, using an engineers trick, whereby the card punch can be used as a card reader since it incorporates a check-reading station that's normally used to verify the correct punching of holes.
card punch with cover off
The card punch made a horrid noise and jammed because the motor was rotating in the wrong direction. All the motors were rotating in the wrong direction, including the drum. The three phase mains was wrongly connected somewhere, although the colour sequence appeared right. The fault was probably in the street, but easily fixed by swapping a couple of phases. That fixed the clock, now we just had to find where the faulty circuit boards might be, and anything else that wasn't working.
Fault finding was pretty easy because there are test points everywhere, in fact, it's all test points. The diagrams are faultless and identify every pin and wire. And the test programs are pretty good at identifying the problem areas. A measure of ingenuity and imagination is still needed to translate the initial symptoms into: which gate or board is faulty, and for me, that satisfying process of deduction was what made the project so worthwhile - along with the things we used it for.
printer with Derek resting
With multiple faults masking each other, it did take some months to commission the machine. I remember the mechanical timing adjustments on the printer were particularly tedious. But once it had settled down, there was a fresh fault only about once a month.
console with cat
The blue sections of panels contain engineer's controls. On the main console, the engineer's section monitors the internal state of sub systems like the DTU, and all the power supplies. From there, the power voltages could be varied, up or down, in an attempt to provoke faulty behaviour before it becomes a problem in normal use.
Some sections were quite a puzzle, because it didn't go wrong often enough to let us become fully familiar with the circuitry. It could take a day to pin down a subtle fault in the tape system, but we would learn a lot by the end of it.
An oscilloscope and a voltmeter were the test tools. D.C. and 1MHz are easy to measure. An operator sitting at the console could send specific signals, data values or operations to most parts of the machine, at full speed, one instruction, or even single clock pulses, and the effects be monitored on the gates by a colleague wheeling (demonstrate) this enormous oscilloscope from one bay to another. Sitting with the diagrams open at the console, one would call out things like "see if 8G27-5 goes high when I clear CR2". And then, power-off, and whip out the faulty card.
Having wire wraps everywhere certainly slows things down, and it's worth making some additional tests
Rodney with hammer
( Web masters Note = Boy but that felt good ! )
to confirm where the fault is, before getting out the special tools. There are rubber strips with a row of holes that hold a card in place temporarily for testing called "quick wraps".
This is an ideal, engineer-friendly, scale of machine. We replaced individual faulty transistors when there were no spare circuit boards of the right type. You couldn't do that with an integrated circuit, or even TTL logic.
printer index wheel
The mechanical parts are nice to work on too, being well made with heavy castings and oil filled gearboxes. Lovely stuff. Having unbuffered peripherals is annoying, but does keep the price down. The main program has to loop and wait for the printer barrel to rotate to the next letter that you want to print. This can be concealed in modular routines but is hardly efficient. It's quite a marvel that it works at all, as evidenced by the measures taken to achieve the clock rate of 1MHz. The many capacitors and coils that populate the circuit cards are there to help these specks of germanium switch at these speeds, -
and do so reliably, when attached to, perhaps, 20 feet of wiring. Apart from what came with the machine, spare parts came from dismantling other 1300's wherever we could find them - with our troop of friends and a big lorry. As machines approached the end of their life, maintenance must have suffered, for we saw a lot of quick-wraps in place permanently.
standing on frame + hacksaw
There was one machine in Bridgewater, and a multitasking 1302 near Hastings with one inch tapes. Another was at the office of the Official Receiver, Tintagel House on, I think, Millbank, - where we were told, a IRA bomb on the window ledge did nothing more than dent a panel, and where, on another occasion security had insisted on watching while a drum that needed changing was dismantled, and the magnetic oxide washed off, in case it contained any sensitive information.
In this way we were able to bring it up to: 6 half inch magnetic tape drives and of course the full compliment of two thousand words of 48 bit (plus 2 parity bits) magnetic core store, and add a couple of extra drums. A full set of spares was put by, and we actually swapped the card reader when we were unable to keep the original running. We added an 8 track paper tape reader and punch, and picked up a variety of punch card preparation machines,
Adrian sitting at a desk card punch
and verifier, with real keyboards instead of a finger dance (demonstrate). Parts get all over the place. The BBC props department used to have a console, it turned up in Doctor Who and other productions occasionally. Any video clips of these appearances would be appreciated. Modifications are easy to make on the 1301. There are gaps in the instruction codes that can be used. You just have to add some gates to decode the instructions and do something interesting. An approved modification called the "black box" involved installing several racks of boards and lets it drive a 1900 series tape deck. This was used to help systems that were converting their data.
One of these was in Glasgow and had quarter inch tapes which involved a two stage transfer via London & Manchester Insurance’s mix of quarter and half inch tapes. The 1302 had a 1900 type interface as standard, but as these systems had all been scrapped, it was left to Galdor to convert the final tapes from several 1301 sites into 1900 format.
IAS barn door
The famous sterling arithmetic instructions are achieved by messing about with the carry of overflows from one digit to the next, as indeed are the decimal instructions. Since the national currency had been DECIMATED, the sterling instructions were redundant, and we made some alterations to the mill to treat those commands as - plain 48 bit binary which was much more useful to us.
The audience at The Science Museum looked horrified at our wrecking a quirky feature that characterised the 1300 series, but relaxed when I continued with - It's all documented, and can revert to sterling by operating a switch in one of the bays.
Modifications were always built properly in 1300 series technology wherever possible. The switch selects binary, sterling or hex, not decimal. We hear that a similar approved modification was installed on some other 1301s to treat the sterling commands as decimal plus a patch in “initial orders” to check the current switch position. We added a "56" (I think) instruction to index or modify other instructions, ie. it adds the bottom 6 digits of the IAS location to CR2. This was a big help in many programs. One practical change was adding some buttons on the console to clear an entire register in one go, without having to dial up a lot of zeroes first. This involved attaching diodes to the back of the dials to form simple pull-up gates. We got the idea from the way the “initial orders” button works - it sets some zeroes and ones into the three control registers to perform a drum transfer and jump. In principle, any frequently required number could be entered this way by adding more buttons.
Drum unit close up
Some changes were made to the original fixed program "Initial Orders". This let the operator set some manual indicators to do things like unload all the tapes, and a very big help was the invention of the G card - When a program is being loaded from punched cards, the last card that is read is normally an E or entry card, whereupon the processor stops and displays all-the-ones in CR1 waiting for the operator to set the program running and it also displayed the program checksum so you could check it! Well, if you overpunch the card's E with a 2, the modified Initial Orders only stops if the checksum is wrong. It works because card code E has a numeric of 5 and G a numeric of 7. To get the program to stop on loading all you had to do was push a chad back into the 2 hole and rub it a bit with your thumb nail.
The modified I.O.s could also sumcheck regions of drum store, comparing a reference checksum punched on the G card, and re-load the program from a library tape on magnetic tape deck 7 if the stored image was corrupt or missing. This can save a lot of time when running a series of programs in sequence unattended, loading each program from tape rather than cards. Only G cards and data need be put in the reader. The ready light for deck 7 was changed for an illuminated push button, so we could put the library on-line without walking across the room.
Another trick was improving the multi-tape file sort program by running a pre-pass of drum based record sorting, processing the keys and pointers in IAS with a bubble sort that made a characteristic whooping noise on the speaker. Small files could then be sorted entirely on the drum, and larger files on tape in a fraction of the previous time. We never understood why such obvious things were not provided by ICL as standard. My colleague Andrew Keene can take credit for much of this sort of trickery.
The loudspeaker gets a click each time a test-and-jump instruction succeeds. One gets to know the normal and abnormal sounds of each program. Inevitably, several music programs were popular.
The original 1301 clock signal generator circuit was improved with a phase-locked loop that proved more stable. I think it uses the 555 chip.
The 1301 1Mhz master clock signal is derived from the currently selected drum timing track via several circuit boards of variable frequency oscillators and counters. Its ability to track the drum and produce a good timing signal reliably was often questioned. The whole lot was replaced with an oscillator and phase locked loop controller on a single IC that had recently become available. A few experiments showed that it would happily work upside down with its ground strapped to the -6.5v rail, more to the point it could track the drum at virtually any speed. The test being to turn the drum motor power off and see how long your program carried on running.
We added an electric typewriter and a keyboard of sorts. This was used to do simple things like show the label details of a magnetic tape. The interface used TTL logic which has compatible voltages but a positive ground, so it had to be built "upside down", and floating, to attach it to the 1301.
The processor stops when it needs to tell you something, and we weren't going to stand over it all the time. So we ran a line up to the house, isolated with a relay, to flash lights whenever it was lonely and in need of attention.
It got quite hot sometimes, and I quite wanted to add an instruction that could be triggered by temperature to release the door panels and flap them about a bit. As it was, we fitted a chimney or two in the central units and a big extractor fan in the roof space. I wanted to get it to control the model train that ran around our flat.
The commercial three phase electricity tariff gave us cheap power overnight, the main reason we ran at night (apart from being students) was the temperature. A nocturnal lifestyle suited us, but confused our clients. We had occasional visits from The Amateur Computer Club and a few schools, but most activity was with customers who appreciated our low overheads.
We wrote programs to look after club and society membership lists. It was called "Clubsoc" and used by some 35 organisations eventually. Nowadays they'd probably do it all themselves with PCs, but in the mid 70's a cheap bureau service was ideal. It started with London Village, and went on to include the UK administration sections of Amnesty International and of Friends Of the Earth; Town and Country Planning Association; National Council for Civil Liberties; the Legalise Cannabis Campaign and Vole Magazine to name just a few. The data was punched from forms onto cards by a separate bureau, and we would produce lists and mail-out labels, usually monthly.
Processing typically was reading punched cards onto magnetic tape, sorting the records into member number order, merging these transactions with the main tape file, and printing various reports and lists, mostly in an automated sequence. Records for resigning members were not deleted immediately but were carried forward for one generation to the new master tape, so that all the reports including deletions could be printed from the same tape.
We also produced printed bio-rhythm charts for a client, and later on, processed the complex administration for the Southern Counties Cat Fanciers shows.
console and me standing
An attempt was made to generate schools' timetables by machine, but the complexities of split sites and individual pupil's requirements meant that this project was not a real success.
The programs were written in machine code and assembler (NPL and MAC). Initial Orders’ relativised loader is an excellent way to combine separate machine code modules, and was a good training in how language compilers ultimately work. We adapted the assembler to use symbolic names and comments.
Writing self-modifying code is fine, and wonderfully efficient, while the programmer is in full control, but would not work now in multi threaded systems.
We provided standby processing for other ICL sites, and did a 6 month period of continuous contract operating in two12 hour shifts at Liverpool Victoria Friendly Society. Their machines were "Samantha" and "Arthur", which suffered a form of Y2K crisis when insurance interest rates reached 10 percent, suddenly needing 2 digits. We also used the G card, and persuaded them that Initial Orders always had this ability, but they were not so sure about the origins of the ability to run a sequence of programs from tape unattended.
We did enough like this to keep the business running, and kept on buying scrap machines, progressing through the 1900 series, - which contained useful amounts of reclaimable gold. We used Cobol and George 3 and moved to bigger premises. I left Galdor when the fun went out of it, and now I provide software support for veterinary surgeons, and am looking for more work if anyone's listening.
In conclusion, the 1301 is of a size where it feels just possible that one could get to know everything about it, hardware and software, including time sharing. Somehow, that's important. It feels uncomfortable using modern PCs and stuff, where one only has the roughest idea of what it's up to.
I can see in the audience, some of the people from Galdor. Perhaps they would fill us in with anything that I've missed.
We handed on the machine in June 1977 when the space was needed for other things, and the next speaker, Roger Holmes, will take up the story. Working around "Flossie" was one of the most satisfying things I've done, and much credit goes to the designers.
Copyright Stuart Fyfe April 2000