A Ferranti Computer Group was set up in 1949 in Moston under J D Carter, Manager of the Instrument Department, to make the computer for Manchester University. In 1951 the Group began to make and sell computers and continued until September 1963 when the business was sold to I.C.T. From July 1951 until the middle of 1963, sales of computers and ancillary equipment, together with computing and maintenance services, amounted to £24½ millions. Total deliveries, however, were only £14½ millions and outstanding orders at the middle of 1963 stood at £9½ millions. In Appendix, Sales of 102 computers are listed. One of these orders - Orion for G.E.C. - was cancelled and in addition several Argus computers were sold by the Computer Sales Department. The Argus is not dealt with in these notes because this part of the business was not sold; nor are the military computers.
During the 12 years techniques changed drastically. The first computer was made with standard sized valves, the next three with miniature valves; transistors and printed circuits were being introduced from about 1956 on an by the time the Department was sold to International Computers and Tabulators integrated circuits were being used.
Until the preparations for War brought necessary changes, the Ferranti Company had three parts, concerned with Transformers, Meters and Instruments. The Instrument Department was often the medium for including the Radio Department, on which was based the electronics work which led to computers.
Rearmament and the War brought many contacts between the Government Departments and Ferranti Ltd, Among these were Dr F C Williams, then at Telecommunications Research Establishment, Malvern, and M K Taylor who was working on the secret I.F.F. equipment.
Another wartime contact between Eric Grundy, who was then managing the Ferranti Instrument Department, and the Defence Departments was Dr (now Professor) Arthur Porter, who was monitoring a contract given to Ferranti. At the end of the War Porter developed a keen appreciation of the need for civilian industry to make use of the experience gained during the War. He subsequently worked for the Ferranti Company in Canada before returning to academic life.
In 1946, F C Williams, with A M Uttley, read a paper to the I.E.E. on the Velodyne, an electro-mechanical system in which a speed of rotation is held closely proportional to an input voltage by feed-back methods. The aim was to determine the position of an aircraft relative to its starting point by integrating Northerly, Easterly and upward movements relative to its starting point. This machine could be applied to the solution of simultaneous equations and many simulations were made.
On 16 January 1947 Arthur Porter commented on a paper read to the I.E.E. by H A Thomas. Porter said there were three reasons why electronics could be of great advantage in the industrial field:
In a paper to the Measurements Section of the I.E.E. on...., Dr Porter and Major Stoneham considered the problem of servo design when input information is only available at discrete intervals. They introduced methods of handling the problem by using cascaded integrators to generate polynomials.
In a footnote to this paper the authors pointed out that the basis of the generation of a single independent variable from tabulated data "can be regarded as equivalent to the method of interlaced polynomials, comparable to the method of interlaced parabolas used by actuaries." One of the most interesting features of electronic computers was the way they brought together, in direct relationship, workers in engineering and in numerical fields of accountancy, insurance, etc., whereas previously such relationship had been rare and mathematical methods in the different professions had grown up largely independently.
Eric Grundy was interested in Arthur Porter's ideas and talked to him, thinking then in terms of electronic control of industrial processes. Porter's advice was that Ferranti should engage Dr Dietrich Prinz, a refugee who had come from Germany in 1936. "DP" was to become the one man who worked in the Computer Department throughout its life and who left his firm impression on a number of aspects of the work and on all who came into contact with him. On coming to England he joined the G.E.C. Research Labs, at Wembley and worked on mercury rectifiers and transmitting valves until, along with many other German citizens he was interned in December 1939. Eighteen months were then spent at a series of camps at Lingfield racecourse, in Devonshire, Canada and the Isle of Man. He was then put in the Pioneer Corps for 10 months. Glasgow University tried to get. him released to lecture in Physics but the War Office would not let him go. Later a former co-internee recommended him to the Bowen Instrument Co. and he was released. He then developed his own servo theory and in 1944 published a paper in the Journal of Scientific Instruments. This led to an invitation to join the Servo Panel of the Ministry of Supply under Arthur Porter. From this contact came the invitation from Grundy who was looking for interesting instruments to develop. Prinz therefore brought to Ferranti a former assistant, McKennell, who designed in detail, under DP's direction, the Ferranti Viscometer. The ideas generated by the I.E.E. papers now seemed to have a man to exploit them.
Grundy asked Sir Vincent de Ferranti to sponsor a study of automatic control from general company funds, but this was refused, and Prinz was employed on a study of radar display for the War Office, which could deal with 250 targets and put on the display about 10 items for each target. The number of targets was reduced to 100 but before the work was complete the crisis of the Berlin Airlift was approaching and the specification was modified, the number of targets coming down to 25.
In 1948, still determined to explore the potential of computers, Grundy sent Prinz to America to see what he could learn about computer development there. Professor Hartree, formerly of Manchester University, where he had developed a large Differential Analyser, was then working in America at the Bureau of Standards. Arthur Porter wrote to Hartree asking if Prinz could be given the opportunity to look at the work being done on computers in America, Hartree considered that Prinz's nationality would make this impossible, but nevertheless DP was sent to see what he could do. He seems to have found his German origins no bar to seeing a good deal of the work being done in the U.S.A. and among others he met Eckert and Mauchley who were working on the computer which was to be the first Univac. There was a nice and often repeated story that it was in America that Prinz was told to go to Manchester University if he wanted to learn about computers. However, Prinz has now destroyed this piece of early mythology; he first heard about Williams' work in the more usual way at an I.E.E. lecture.
Grundy discussed with F C Williams at Manchester University the possible source of finance for building computers. "PC" suggested he should talk to Sir Ben Lockspeiser, Scientific Adviser to the Ministry of Supply. The opportunity came when Sir Ben was visiting the University on the invitation of Professor (now Lord) Blackett and Grundy was sent hastily back from a visit to London to see the computer work Williams was showing them.
The story of this first computer, according to Dr B V (now Lord) Bowden, was that during his time at T.R.E., Williams had developed his technique of storing charges on a cathode ray screen and from this had developed a means of representing digital numbers. At the end of the War T.R.E. became interested in the possibility of making a computer and formed a committee to study the problem. Williams was a member of this group and was asked to develop the circuits which would be required. In due course the other members visited him and said they were making progress with their design plans and how were his ideas on circuits developing? Williams said he thought they were coming along satisfactorily but of course he had had to make a computer to test them and would they like to see it? This was the precursor of the engine which Grundy was called to see with Sir Ben.
F C Williams explained that his principal difficulty was to make a magnetic drum to serve as a large backing store for the fast C.R.T. store. Sir Ben Lockspeiser thought that it was important that a fully engineered version of the Williams computer should be made and that he could find the money. He later explained that at that time he could find money more easily than men. From London he wrote to Ferranti telling them to proceed.
There was a slight hitch when Carter, with Bob Hooker, the Ferranti Governments Contracts manager, went to see P S Barton at the Ministry of Supply, as they thought to collect the formal contract. Barton said he could not give this: tenders must be issued and competing quotations assessed. Hooker then showed him Sir Ben's letter and told him the work had begun. Sir Ben never showed any sign of repentance over the cavalier treatment of his contracts department.
About this time Sir Ben was transferred from the Ministry of Supply to take charge of the Department of Scientific and Industrial Research. He had seen that computers would become very important and wanted to have three made and before he left the Ministry he applied to the Advisory Council on Scientific Policy. There Sir Henry Tizard ruled that the promotion of computers should be done by Sir Ben from his new post as the Secretary to D.S.I.R.
This led to a setting up of a Committee under the Chairmanship of Sir David Brunt. Sir David had served in France as a Meteorologist in the first World War; he had been head of the Meteorological Office and, in 1936, the non-playing captain of our international Gliding team. When we dealt with him he was the Professor of Meteorology at Imperial College and Physical Secretary of the Royal Society. The Brunt Committee was to put on a sound basis Sir Ben's desire to promote computers.
In 1948 Sir Stafford Cripps, then Chancellor of the Exchequer, set up the National Research Development Corporation to exploit the many new inventions which the War had produced and those which subsequently were produced in Government establishments or at Universities and elsewhere under Government Contracts. The Williams patents therefore came under the control of N.R.D.C. and when Ferranti began to consider the manufacture of further models of the Manchester computer they were dealing with the University and N.R.D.C. as well as potential customers.
In the making of the first computer Ferranti had built up a highly-trained and now experienced team. There was now the danger that there would be no work for them. However, Brigadier Hinds at the Ministry of Supply was wanting a computer for the Armaments Research and Development Establishment at Fort Halstead and gave Ferranti a letter of intent and the building of a second computer began. When I joined Ferranti at the end of June 1951 they had just learned that the sum needed to buy this computer exceeded the amount Brigadier Hinds was empowered to spend. Hinds insisted Ferranti knew they had to have an official contract before the Ministry could be committed and the situation was rather delicate. Fortunately, Ferranti interests fitted in with the strong views about computers held by Sir Ben Lockspeiser and the ambitions of Lord Halsbury at the N.R.D.C, and after a time the Fort Halstead situation was rectified.
The Manchester University Computer was formally opened by .......... on July 12 1951 with a very large attendance of scientists, Government officials and interested business men. It was clear that a revolution had begun, though it was to take some years before schemes first suggested at that meeting were to yield results.
To Ferranti this was the advertisement to the world that they were ready to make and sell computers. We have always claimed that the Ferranti Mark 1 computer was the first to be offered for commercial sale anywhere in the world.
The first sale was made to the University of Toronto through the close association of the Ferranti Electric Company with Professor Watson of the University.
Before continuing with the history of that computer the following pages give a brief story of the technical development of the Mark 1.
When F C Williams showed that charges on the face of a c.r.t. could be satisfactorily regenerated, he made possible:
The first was an essential requirement in a computer, the second was of course very important and the third feature was, many years later, to become a major development. In this early machine the c.r.t. was, in the language of today, a man-machine interface. Numbers were made to appear as a two dimensional array of bright and dim spots representing respectively "1s" and "0s" in a binary code system and these could be used to draw charts or pictures. Prinz made a test programme, which he called the "Hook-Noes" programme which, if the test was satisfied, displayed a column of ticks or, if it failed, a column of "No's". The convenience to the programmes of this facility was clearly recognised, particularly in the early days, when there were frequent questions as to what had happened inside the computer and the possibility of having consoles with c.r.t. screens in rooms away from the computer could be anticipated.
The advantages of the c.r.t. were not enough, however, to compensate for its disadvantages compared with other types of storage, particularly core stores when they came in and the development of c.r. t. storage was not taken up by companies other than Ferranti in the U.K. though it was used and more work done on it in America and in Australia. Interaction between the user and the computer awaited timesharing computers before it became widespread.
Williams was not the first to conceive the idea of using a television-type raster to produce arrays. His contribution was to show how, when charges had been put on the c.r.t. screen they could be retained as long as the power was turned on.
This demonstration was achieved in 1947 and the next job was to use this type of store to give a computer an internal memory and in June 1948 an experimental "baby computer" with a capacity of 32 words, was built. This, though too small to do more than demonstrate the techniques and suggest the promise they offered, was a "universal" computer with a store, as abstractor and a control unit and could solve any problem within its capacity which could be reduced to a programme of elementary instructions, and if enough time could be given it. Although it was a test bed for new techniques and not required to show great speed in performing operations the baby gave an idea of things to come. In a factorizing program in which division had to be done by repeated substraction, 3k million operations were performed in 52 minutes.
About this point the staff was increased. D B G Edwards, who later became Professor of Computer Engineering, joined the group and others working on the project were G E Thomas (now Director of the Regional Computer Centre at Edinburgh) and A A Robinson (now Director of the National Computer Centre).
The c.r.t. store had an accessibility ratio (the ratio between the time taken to read the contents of a chosen address in the store and the time between asking for the information and the end of the reading process) of unity. To make an economical system with the very large store the programmers required meant that speed had to be sacrificed and this was done by providing, as a backing store, a large magnetic drum on which digits were stored as magnetic spots, the direction of magnetisation indicating "1" or "0". These magnetised spots were arranged in tracks around the circumference, each track holding 2560 binary digits, the same as the contents of a c.r.t. store in which these digits were held as two rasters each with 32 lines of 40 digits. The experimental machine now made contained 5120 digits of electronic storage and 40960 of magnetic storage.
The third stage of the development was the prototype for the Mark 1, the computer which among those who saw it, is generally regarded as the first version. It was a roomful of racks linked by a dangerous looking network of wire. It was said that when he saw it Sir Ben Lackspeiser asked "Does it work?" and the engineer switched on and, for the first time, it did. The story almost certainly aprocryphal.
The experimental computer had too small a fast store for the convenience of the programmers so the prototype was given a fast store of 10240 digits and a magnetic store of 150000 with provision for 600,000 digits.
In August 1949 Manchester University started passing information to Ferranti with G C Tootill as the Ferranti man on the project. The computer took nearly two years to complete.
The input to the prototype was 5-hole paper tape using a photoelectric reader capable of reading 200 characters a second.
The output was by means of a paper tape punch or teleprinter each capable of working at 10 characters a second. Input speeds measured in hundredths of seconds and internal calculating speeds in milliseconds meant either that the calculations for which computers were suited were those in which a lot of calculation had to be done on relatively few numbers in the computer or accept that the electronics would waste a lot of their time; an expensive waste. Many of the calculations for which computers were expected to be most useful were just those which preceded by calculating numbers which were required in further stages of the computation. These were the problems - involving sub-totals or stages in iterative calculation - for which existing machines were very slow. Punching sub-totals or intermediate products on punched card machines is wasteful compared with holding the figures in electronic storage from which they can be extracted on demand.
The Manchester Mark 1 introduced a valuable facility in the "B-tube" which held a "modifier". In the original machine this held a single number which was added to the instruction or not as the programmer required. It was a device for iterative computation which saved storage space: for example, in operating on a series of pairs of numbers in consecutive addresses the storage of units in the "B-tube" would alter the address portions of the number by one after each operation so that the machine could move down the columns of numbers automatically. The "B-tube" was so-called because the letters "A" and "C" had been nominated for the accumulator and control tubes respectively and there was no very obvious letter which the designers felt compelled to use.
In the first design of the "B-tube" facility it was necessary to use the accumulator in association with the "B-tube" in order to count that the correct number of iterations had been performed. It was then found, by a study of the way the "B-tube" was used that it was possible to write the count number in the "B-tube" without recourse to the accumulator. This saved instructions.
Quite apart from Professor Watson's natural desire that his University should have the distinction of being among the earliest places in the world to have a computer, the University wanted to help the Canadian Government in its share of the cutting of the St. Lawrence Seaway. There was perhaps some political benefit to be gained if Canada rather than the United States did the major part of these calculations.
There was also urgency to have early delivery of the computer for financial reasons. Funds to pay for it became available in the year 1951/52 and it was important to prepare to have it there for a Computer Conference in September 1952 and also to be able to justify paying part of the money in that year. This was unfortunate. The University computer was showing how difficult it was to make a reliable instrument even with designers and manufacturers and the site within a few miles of one another. We now had to install the computer 3,000 miles away from the base and with only a few people available who understood it; and to meet the peculiar obligations of Government financing, which seems to take no account of practical problems in technology and transport, it had to be dispatched in stages so that there would be no opportunity to test it fully assembled except on site. Perhaps it would have been better if the first delivery had been made, while duplicate equipment was used to make the complete system and the original delivery then returned. It would have been like the old problem of transporting wolves and sheep over rivers but it would have saved later trouble.
The St. Lawrence Seaway job however was an opportunity for N.R.D.C., and particularly for Christopher Strachey who was lent to develop the program for calculating the effects of the plans on the water flow of the river past the Thousand Islands. Vivian Bowden had sold the computer to Toronto University in 1951; Strachey went over in the autumn of 1952 and in three months the computer had done sums which it had been estimated would take clerks with hand machines 20 years. We do not seem to have been told how many clerks so we cannot calculate the increase in productivity. More important was the saving in time.
There were considerable teething troubles with FERUT. The engineers had not been allowed time to get it all working before dispatch to Toronto. Everyone expected some trouble from the electronics, which were new, though there had been thirty years experience of using electronics, often in most unfavourable conditions. But, as in Manchester, it was the "old fashioned" electro-mechanical equipment which gave most of the trouble. Making a magnetic drum had been one of the biggest problems for the University Mark 1 and it was repeated for the Toronto machine. There had been trouble in Manchester in making power supplies which would eliminate the effects of sharp changes in the mains voltage and again in Canada the power supplies proved a problem. These had been supplied by Campbell and Isherwood and after a series of failures the generator had to be stripped down and rewired. The argument grew until Professor Watson wrote to the President of the Board of Trade protesting about the quality of British goods. This put pressure on the suppliers to produce a standby which was ordered from Westinghouse.
The programme developed by Strachey was said to need uninterrupted runs of 5 to 10 minutes duration, though elsewhere a need for only 5 minutes trouble-free work is mentioned. Probably the circumstances calling for the longer time were rare. With a machine which might fail towards the end of one of the longer runs the time spent in repeating was extravagant and changes were made in the program to increase efficiency. For example, a characteristic of the program was that data and computation were sometimes recorded on the same tracks on the drum which, in the event of a computational error meant that all the data had to be read in again.
The third feature of the job was that the operators were inexperienced and had to learn from their mistakes. One report shows that for 8 out of 9 cases calculated one day the answers had to be thrown away because wrong data had been used; another that only 30 minutes useful work was achieved in 7½ hours because of faulty operator work.
The fault of all kinds fell heaviest on the maintenance engineers who might have to search in the computer for all faults, real or imaginary, whether in the computer or in the programs or through operator's procedure. Hodkinson, the chief maintenance engineer, who had gone to Toronto to look after the maintenance, worked for 650 hours between 1 January and 14 March 1953. It was largely because of his labours that by 24 April, 80 per cent of the Seaway project was finished. Professor Watson was well satisfied and although there was clearly much to do to perfect the system the computer had probably already justified its cost. The relative speeds of computations compared with the hand methods was given as on average 400:1 and in one case 1800:1.
The maintenance was handed over to Ferranti Canada, and it had been hoped that business would flow from it. That it did not was partly because its reputation had suffered - other computers, e.g. Illiac at the University of Illinois appeared to be more reliable - or because the input and output facilities of FERUT were limited, too limited for example to satisfy Wiseman of ..... who might have been a customer.
The performance of FERUT was recorded as being 64 per cent for the 1st quarter of 1953, 78 per cent in the second and 79 per cent in the third. However, these figures, though they indicate that there was an improvement in performance, do not give a measure of useful time. The calculation related only to the performance after engineering time had been spent, so the more time spent on engineering the less there was for faults. Also "Idle Time" was included as good time because the computer had "ticked over" without producing a fault.
It was probably a mistake to take the Toronto University order so early when we were unsure of the problems likely to arise. We had not studied what could be done and, more important/ what potential customers wanted to have done with computers. The modifications which produced the Mark 1* were in hand and by 1954 we could have delivered a much more reliable computer. But one has to admit that it was the first order and so must have had much influence in getting us launched, though it did not give Ferranti a high reputation on the North American continent.
The proposal to set up that National Research Development Corporation was made by Sir Stafford Cripps in 1948. It took the Government some time to find suitable people to launch the Corporation and it was on May 10 1949 that Mr Harold Wilson, then President of the Board of Trade, was able to announce the Lord Halsbury had agreed to become the Managing Director when the Corporation was formed. By the end of June the Board was constituted as follows:
Chairman: Sir Percy Mills Directors: Professor P M S Blackett, F.R.S. Sir John Duncanson Sir Edward de Stein Sir Edward Hodgson K.B.E., C.B. Mr W E P Johnson A.F.C, C.P.A, F.R.Ae.S.
Patents which had been taken out by Government Departments and by Universities and other bodies working on Government contracts were transferred to the N.R.D.C. who were given £5M as working capital. They were expected "taking one year with another" to meet their future expenditure out of income. The patents in the Manchester Computer, particularly those in the Williams Tubes, thus became the property of the N.R.D.C.
N.R.D.C. ownership of the computer patents meant that the Corporation took responsibility for dealing with the Americans, particularly I.B.M., who might make claims against manufacturers using the new circuits and components. This was a valuable protection for Ferranti and other companies who might with this help feel able to take the commercial risks.
N.R.D.C. proposed that a partnership should be formed between Ferranti and themselves in which they would provide the money. This was rejected by Sir Vincent who found the borrowing terms too expensive compared with those given by the Westminster Bank. He said that N.R.D.C. would best help by ordering a number of machines to enable Ferranti to keep together the team of expert engineers and to relieve the Company of the risk and cost of stocking the computers. After consultation with the Brunt Committee, N.R.D.C. proposed that they should order four computers, one of which would be for use by Ferranti for research into the problems of programming and applications of computers. A letter of Intent was given in June 1951 and a contract was signed on 3 October of that year.
In brief, the terms of the contract were:
That Ferranti should make four computers, with expedition and efficiency, to be paid for on the basis of cost plus 7% per cent, costs to include all direct and overhead costs properly applicable to the contract, including selling costs, and in addition Ferranti were to receive 2½ per cent of the price as a selling commission as a reward for success in "using their best endeavours to find customers". This contract established a close working relationship between the Company and N.R.D.C.
Although the Manchester Computer aroused a lot of interest and was one of the most, and probably the most, powerful in the world, N.R.D.C. were running a risk. Interest was not easily converted into orders. There were a number of marketing difficulties:
On 25 April 1952, Bowden had to report that we were unlikely to sell to Universities at the moment.
Selling was consequently interesting and often exciting but unrewarding. Time and time again we could only report, after customer visits, that they were showing a keen and sometimes a serious Interest, but no order. Fortunately, N.R.D.C. were bearing the costs and we were able to make a large number of contacts who were later customers for one or other of our computers - or our competitors'. Persistence paid: the record was held by the negotiations with the Metal Box Company. The Chief Accountant and H W G Gearing came to the opening ceremony. Harold Gearing was an old acquaintance, an accountant and a statistician who was immediately attracted by computers, worked hard on each of the models we produced and tried several times to persuade his Company to buy one, but although John Ryan, his Vice Chairman, who was well known to us and a former mathematics don at Cambridge, seemed to support the efforts, caution held them back until 1963 when our twelve years efforts brought an order for Orion.
A sign of change came in the middle of 1952 when a team from the Shell Company visited us. They had been looking at computers in the U.S.A. and at Cambridge and the N.P.L. They heard of a machine at Manchester and, having a day available, came to see us. In November they called for a quotation and shortly afterwards decided to buy a computer. By June 1952 we had been told the Ministry of Supply would want two computers which were ordered in October. So by the end of that year the prospects seemed to be transformed and the N.R.D.C. - Ferranti collaboration was really in being. It was clear that a roomful of electronics, needing elaborate installation procedures and an expensive maintenance staff would only be suitable for a small class of customers and there was need for a smaller and cheaper computer directed towards the commercial market. We had experimented at the University on preparing a payroll and found that the cost by computer was much the same as when done by hand, but computers could also use the wages data for building up cost accounts and statistics.
Professor Blackett was interested in this subject and at a meeting with Bowden and Swann on 17 June 1952, he outlined his ideas for a PAYE computer and at subsequent meetings on June 20 and July 7 and 14 there were talks about a computer for commercial purposes and also about a card-to-magnetic tape and magnetic tape-to-card computer and the design of suitable gang punches and collators. Later, in the spring of 1953, N.R.D.C. suggested they should buy a magnetic tape transport from America and give it to us for experiments. This was opposed by Pollard who had been made Chief Engineer of the Computer Group and regarded it as a "waste of public funds" and little progress was made. Bennett made a design proposal for a "wages machine" based on a magnetic drum - somewhat like the computer designed in Holland which later Hugh Ross wanted us to consider and which was taken up by S.T.C. as the Stantec Zebra - but Pollard did not accept John Bennett's ideas which were later submerged under the plans for collaborating with Power Samas.
Professor Blackett accepted the arguments we put forward that a wages machine should be able to do quite a lot of other work. Our Wages programme was more complex than had been expected. It revealed clearly one of the essential features of commercial work which would always make programming it difficult. There were rules which had to be considered in every case though they only rarely applied (e.g. the entitlement to a return of tax). There were arbitrary items such as temporary bonuses to employees who lost money through changing jobs which lowered their piece-work earnings; the discontinuities caused by the steps in the tax tables were much store trouble to a computer than an elaborate mathematical formula would have been. Programmers were soon to find out that although commerce mostly used very simple arithmetic human activities were represented by complicated procedures.
Scientific work was more promising and we were soon busy with the problems raised by Shell, the Government Departments and the University of Rome where Professor Picone early showed interest. We became so busy with customers of this type that we tended to leave the accounting type ones until we had punched cards. This led Lord Halsbury to tell his Board that Ferranti had tried to do commercial work, had "fallen down on the job and handed it over to Powers". There was too little discussion between N.R.D.C. and Ferranti. The Ferranti management too thought that customers should flock to buy. After we had sold about £1 million worth Mr W A C Bass the Director who kept an eye on the Computer Development, said "you cannot sell these machines" and agreed that Powers should do it for us. Unfortunately the salesmen in Powers showed little interest; commissions would not be earned without a lot of preliminary work.
The Appendix "Some Programmes written for Ferranti Computers" gives a very brief account of a number of jobs which were programed on the Manchester computer during the first four years by Ferranti staff. A large number of jobs were also done by the University and other researchers who used the computer.
Over the last few years a great many problems have been solved on the Manchester University Computer and other Ferranti machines. The following is a description of some of the more interesting programmes that have been written.
A programme was written for the calculation of the critical rotational speeds of a rotor system corresponding to the neutral periods of lateral vibration of the system.
This problem should not be confused with the above. This concerns the torsional vibrations set up in driving machinery, and the basis of calculation is the Holzer method.
Several programmes have been produced at Manchester University for the analysis of the forces and couples acting upon rigid frameworks.
Two and three dimensional ground to air missile studies have been carried out.
The solution of the differential equations of the ballooning course of a fine yarn during ring and cap spinning were achieved with a computer.
This programme concerned a problem of importance to power engineers, namely the determination of phase angles and amplitudes of currents and voltages at points of a passive electrical network when the powers supplied to or extracted from various nodes of the network are given.
A useful tool in the analysis of waveforms found in experiments is the application of the methods of Fourier transformation. When the number of points involved is large it becomes impossibly tedious to use hand computation, and it is then that a digital computer is especially useful.
From the solution of the Poisson and Helmholtz partial differential equations based on observations at a number of meteorological stations weather forecast charts have been produced on an experimental basis. The results, when compared with actual observed weather, have been encouraging.
A programme was written to calculate the grid co-ordinates of points fixed by the method of traversing. The experiment showed what features would be especially desirable in a computer built solely for this type of work.
From observations on the diffraction of X-rays by crystals it is possible to derive a 3-dimensional map of the distribution of electrons in a molecule, and so determine the exact spatial arrangement of the atoms composing the molecule. These calculations are very lengthy and the use of the computer has proved to be of enormous aid.
A computer has been used to aid lens design. The programme is able to improve standard lens systems in a completely automatic manner.
The purpose of the programme was to compute and print a set of partial regression Coefficients, but it was extended to calculate the standard errors of the regression coefficients, perform t-tests of significance and where necessary to compute a new set of coefficients omitting those variables whose coefficients are found not to be significant.
A programme has been developed to deal with the solution of partial differential equations by the methods of relaxation. In order to make use of the programme, the problem for solution must be expressed in finite difference equations in a code which it then interpreted by the computer.
A number of programmes of an interpretative nature have been written. Special attention has been paid to matrix manipulation schemes, where the various manipulative processes are represented by some simple code, which is translated by an interpretative programme into the normal machine code.
Such programmes have also been devised to test programmes written for machines using different codes.
An optional function with the Ferranti Mark 1* computers is a random number generator. However, programmes have been written to produce locally random numbers for problems such as traffic control and random noise in amplifiers.
Many programmes have been written to aid the engineer in locating faulty components in a computer. The purpose of these programmes is, wherever possible, to narrow the area of search for faults. In this way the computer can be an invaluable tool for maintaining itself in good running order.
In order to reduce the amount of machine time spent on the location of errors in newly written programmes, several special programmes have been written. These programmes are designed to print out, information about the contents of stores and registers at any desired point of the programme being developed.
Several programmes have been constructed to make coding simpler. These enable those who cannot afford time to learn the normal computer codes still to use the computer for their problems.
A demonstration programme was written to show that a computer can be used successfully for producing payslip. Information was input on paper tape and the output was done by a High Speed Parallel Printer. The experience gained from this experiment has helped to indicate lines on which improvements can be achieved.
A machine was used to demonstrate the calculation of Board of Trade Index Numbers. These figures are produced on a monthly basis for all trades and serve to indicate price trends. Input of information was by paper tape and output by the High Speed Parallel Printer.
The transportation problem may be stated as follows. Given a number of sources and destinations for goods, to determine how the goods should be dispatched to make the cost of transportation of an order a minimum.
The transportation problem is one special example of linear programming. Programming in this sense should not be confused with programming in the sense of organising machine instructions. The thoery of linear programming has wide application in all fields, but a particular example may suffice to illustrate its use. A particular problem tackled was that of finding the diet of basic foods which would provide adequate vitamin, mineral and energy requirements for the lowest total cost.
Demonstration programmes were written to prove the capability of electronic computers to do the work of an ordinary life assurance office. Since paper tape input was the only available input medium the programme served to evaluate the principles and made it possible to make time estimates for a computer using magnetic tape input/ output units.
The demonstration was concerned with the printing of policy records from a tape file, the production of agents' monthly lists for any required renewal month, and the valuation of all policies in the summary form of the 4th Schedule to the Board of Trade returns.
A programme was used to test various methods of reducing the losses incurred in meeting orders for rectangular pieces of material. The results proved encouraging and suggested a considerable saving in wastage might be achieved by using such methods.
The rapid analysis of production plans to estimate relative profitability has been demonstrated on a computer The problem is to determine how much of each raw material and each intermediate material is required to produce given tonnages of a final product.
These notes refer to work which has actually been programmed and carried out on a computer. Other commercial jobs which have been analysed and for which programmes could undoubtedly be written include:
The story of the relationship between Ferranti Computer Group, as it was then, and I.C.T. began with a need to be able to use an established method of recording data in conjunction with the Mark 1* computer. It progressed to a link-up with Powers Samas Accounting Machines Ltd which ended in frustration to both companies. Powers diverted effort without getting the computer they were looking for; Ferranti sacrificed business and diverted a much higher proportion of their limited effort and did not get what they so badly needed by 1958, a saleable successor to the Pegasus computer.
By contrast, Elliott Bros established a link with the National Cash Register Company which was, apparently, very successful. The most important differences between the two exercises were (1) that National Cash had customers who were needing computers and they were given a chance to sell a product which was already being marketed, whereas the Ferranti-Powers team set out to make a new product and meanwhile had to suffer commercial and technical restrictions; (2) both Powers and Ferranti had computers of comparable size which put their sales teams in competition, whereas the sales efforts of N.C.R. and Elliott were complementary. It had not been that there would be serious competition between Ferranti and Powers because in the early days there was supposed to be more difference between scientific and commercial computing than later proved to be the case.
To put the coded programs and data into the Manchester Computer, the University used punched tape. It was cheap and sufficed to prove the engineering design. Cambridge University also used paper tape while for the ACE the National Physical Laboratory staff used the British Tabulating Machine Company's equipment. They were already processing a lot of technical information and their data was held on cards. In America Eckert and Mauchly, after studying the enormous amount of information used in commercial calculations, decided to go straight to magnetic tape, taking on a very difficult piece of research and development which was to be of enormous benefit to the whole computer world but which was sorely expensive to the pioneers.
The Ferranti management realised they would have to be able to use punched cards and also that the B.T.M. Co and Powers Santas each had a great deal of commercial systems knowledge and very many existing customers who would be tied to them for quite a long time. Discussions therefore took place in 1950 (?51) between Messrs Bass, Grundy and Carter and Dr Bowden and the B.T.M. directors, but these talks were broken off when it was clear that B.T.M. would insist on having sole responsibility for marketing. Ferranti were reluctant to give away this part of the control of their business. They probably saw more clearly than the punched card experts the importance of the large electronic memory as a means of storing the immediate results which arise in scientific computation and which need to be quickly available for the further stages. Punched card men seemed to expect to find the greater value of electronics in helping to overcome the slowness of multiplication.
Ferranti do not seem to have contemplated formal discussions with Powers Samas at this time. They used B.T.M, machines and the two companies had known one another for a long time. However. Pollard had talked about computers to the Powers engineers who were making a small machine with punched card input and output (the P.C.C.). This, had it proved more reliable, could have been the beginning of a range of computers which might have given Vickers control of the computer industry. (The same might be said of English Electric if in the early fifties they had moved more vigorously to promote copies of the ACE which they had built to N.P.L. designs). Pollard saw the opportunity given to Powers and shortly after I joined Ferranti he approached me with a suggestion that he and I should leave the Company and go to Powers. I did not consider the proposal and I am sure that had I done so I should have judged it unfavourably.
Pollard and others who later proposed or supported the idea of an exclusive relationship between Ferranti and Powers did not seem to understand why the latter company's equipment was not commonly used for scientific and technical experimental work. The Powers card is "read" by pins passing through the holes punched in the card in positions which represent numbers or symbols. The connection box, which links the matrix of pins by a number of Bowden cables to the computing engines, had to be specially made for each job. The system was therefore inflexible. If a change in a job required a re-arrangement of the connections a new box had to be made which would normally take weeks. The system could be claimed to have two main advantages: the electrical contacts made by the pins were very certain and the fact that once planned the connections could only be changed by a replanning procedure and further engineering meant that the operators could not, whether intentionally or accidentally, make changes. For work such as insurance or the population census or similar jobs, these were advantages. They enforced very careful planning and were free from interference, however well-intentioned. In technical or scientific experimental work, however, they produced very grave difficulties. The experimenter often needs to change his computational approach as the first attempts produce answers which are insufficient and the Hollerith method permits these changes to be made by the operator at any time by altering the connections in a standard plug board. In general, scientific and technical groups tended to use Hollerith cards. The changes needed to adapt the Powers system to get the benefit of the Hollerith plug board could perhaps have been made and indeed may have been contemplated, but immediately the Hollerith cards were the more important to Ferranti though we knew we should very soon want to seek business with insurance companies and the Population Census was a candidate for computers. (I had been a member of the Government Committee planning the requirements for the 1961 Census and so had a personal interest in wishing to get this business and had taken first steps on joining Ferranti, but had to slow down as relations developed with Powers who regarded the Census as one of their prizes).
In the summer of 1952 it was agreed at a meeting in the office of Mr (now Sir Cecil) Mead, Managing Director of B.T.M., with Kenneth Elbourne, his Sales Manager, and attended from Ferranti by Messrs Bass, Grundy, Carter and Swann that Elbourne and Swann should meet to work out a basis for collaboration. They held two meetings which resulted in an exchange of letters: (1) B.T.M. agreed to help Ferranti to put punched card input/output on their large computers, (2) they would not help Ferranti to make a small computer because they were developing one based on the design of the APEX computer made by A D Booth of Birkbeck College and (3) it was recognised that Ferranti would make smaller computers and B.T.M. larger ones, but agreed that this problem should be left for the future. I told Elbourne I was due to talk also to Powers Samas about input/output equipment to our computer and asked if there would be any possibility of talks between the two punched card companies. Elbourne said there was no possibility of this, they were in fierce competition. During the period of our negotiations with Powers the force of computer competition compelled the two punched card firms to come together. But in 1952 we had to deal with them separately so Elbourne and I arranged to get the engineers of B.T.M. and Ferranti together.
I had known Elbourne since before the War when I was helping to reorganise the Sales Statistics in I.C.I., using a large centralised machine roomful of B.T.M. equipment. I had also had considerable dealings with him during the war because in 1940 my Director and I had helped him to persuade General Edgecumbe, the Director of Organisation to put in B.T.M. equipment as part of the planning for the demobilisation of the Army. (There was some reluctance in the War Office which may be explained by an incident which took place on the day of the opening of the Manchester Computer. Ronald Michaelson, who had joined B.T.M. to help in their studies of computers, said to General Bednall, then Paymaster in Chief, that he ought to install punched card equipment and "catch up with the Americans". The General replied that on the contrary they were in advance of the Americans, they had put in Powers punched cards and already thrown them out again). The B.T.M. installation fully justified itself as an administrative instrument during the Wat and certainly smoothed the demobilisation process.
The B.T.M. objective was to get electronic machines which could easily be seen as substitutes for existing electro-mechanical ones, particularly to make multiplication faster. At a party given to the Commonwealth Statisticians by the Board of Trade in 1951 Elbourne had said that he saw the future punched card machines as "having electronic insides". Both punched card companies look on office machines quite differently from Ferranti - they saw them as machines for economising manpower i n routine operations, while the University and Ferranti staff were more interested in their use for doing calculations so big they had so far been virtually impossible, and doing technical calculations in research and development quickly.
Our experiments had suggested that though wages calculations could be done more quickly on computers than by current methods they were unlikely to be done more cheaply except in a few very large organisations. We knew the troubles caused by machine failure and it seemed to us that it would be a little time before we could get the reliability which an economical take-over of clerical work required. The difference in attitude came early to me one day talking to Warwick Deacon of Powers who said "Do you mean to say that when you sell a computer you tell the customer he must also increase his staff?" When I said this was necessarily so since they had to have programming and maintenance staff he was astonished and said "I don't know how you ever sell a computer". B.T.M. were probably more conscious of the importance to them of work arising in scientific and technical fields. Certainly the association with Elbourne seemed promising. Before we held our formal meetings he had asked for my help at Nielsen's of Oxford whose business he was in danger of losing to I.B.M.
The B.T.M. talks made no headway. They began between Womersley (a mathematician who was formerly with the N.P.L.) and Dr King, head of engineering development, on the B.T.M. side and Pollard, Bennett and Swann for Ferranti. Unfortunately Pollard was opposed to talks with B.T.M. because he wanted to collaborate with Powers. Dr King was cautious, no doubt feeling unable to speak too much about what they were doing, and Womersley was far from clear. He was perhaps in some difficulty. He was on a fixed term contract which had not long to run and which in fact was not renewed, and he was also perhaps a sick man. He did not live long after leaving B.T.M.
There is some similarity between Ferranti experience in dealing with B.T.M. and that of N.R.D.C. who also found their conversations faded away. No doubt B.T.M's position was difficult; it was fundamentally powerful but weakened through failure to get to grips with the large computer early enough. They could so easily have given ourselves or English Electric more than they got in return. The Ferranti-B.T.M. talks finally broke down because of the publication of a Ferranti-Powers arrangements and an incident hardly connected with business.
Discussion with Powers Samas began with a meeting at Kern House in July 1952 when Messrs Nash and Johnson met Pollard and Swann. Fred Nash was the Managing Director of Powers Accounting Machines (Sales) Ltd and Johnson was his Chief Engineer in charge of Research and Development. The Powers Samas Company was owned by Vickers who had bought it from the Prudential Assurance Company to provide work for their Fire Control Department. It had been founded in the U.S.A. by an engineer who worked with Dr Hollerith but broke away and had to find a way of reading and punching cards which would not infringe the Hollerith patents. A great deal of ingenuity had gone into the system but developments had given competitive advantage to the Hollerith system. The greater flexibility of this probably had a lot to do with it and, in the modern world, electricity has advantages over Bowden cables as a means of communication. Also, the success of I.B.M. in building a big business based on the Hollerith system added to B.T.M's competitive advantage over Powers though now both companies were threatened by the American giant: B.T.M. had terminated the agreement with I.B.M. under which they divided their spheres of interest.
Powers, perhaps, saw more clearly than B.T.M., that computers were a means of capturing markets in which they had not been strong. It was agreed that talks between the engineers should take place and Johnson immediately got to work on equipment to feed Powers cards to the Mark 1* Computer. This was not successful, however, and no sales were made. It was never regarded by Ferranti as a product for them to sell though an abortive attempt was made to sell it to A V Roe.
Nash invited Swann to meet his Chairman, Robert Wonfor (now Managing Director of Roneo Ltd) in November 1952. Wonfor put forward the suggestion that Powers might buy a "part of the Ferranti computer" to make a small and simpler business machine. He emphasised the need for secrecy, particularly that B.T.M. should not get wind of the project. This was not difficult. We were not talking about small machine problems to B.T.M. It was a mistake, looking back, that we ever got involved in this project; it would have been better to stick at this stage to the original idea of just getting equipment to work with our big computers. But the prospect of being able to manufacture a small machine on a large scale for a company rich in cash which would take all the responsibility and costs of selling and maintenance was very attractive to the Ferranti management.
The talks were based on two principles:
The protracted discussions failed to produce that promised new computer though a Powers card input/output system was added to the Ferranti Pegasus computer in 1958 and announced under the name of Pluto. The Powers sales team in the course of 5 years produced no orders but by their intervention caused the cancellation of two firm letters of intent and frustrated other prospects. The restrictions put on the Ferranti sales staff did not prevent them from getting orders to keep the factory occupied, but if their hands had not been tied their efforts would have cost less and many more sales would have been made.
The Powers organisation was a two-level structures a production company which designed and manufactured machines to specifications and prices which had been agreed with the Sales Company. Marketing knowledge was very considerable. The close relationship which any monopoly supplier has with his customers is strengthened when he has to supply maintenance facilities, and as the punched card companies also took responsibility for planning systems they could see clearly how and when to intervene with a good chance of selling new equipment.
Powers also kept a very close watch on all the activities of their competitors. The digestion of all this information to specify new products was in the control of Whitwell, an actuary with long experience in computational work who had worked with the famous L J Comrie of Scientific Computing Services. Comrie had devoted much of his time to the ingenious use of business machines in scientific calculations.
Whitwell was an able mathematician who occasionally lapsed into bluntness which could upset relationships. I had first come across him when he had been asked to be an examiner for the Association of Incorporated Statisticians, of which I was a member of Council. He came to see me in Manchester over an upset he had caused there - which I never fully understood and could probably not have helped with, but later experience made it very easy to guess the sort of thing that had happened.
Later in a joint meeting with a prospective computer customer he was to make a remark about his "having to keep the Ferranti designers' feet on the ground" which very much upset Welchman, his opposite number on the Ferranti staff.
Gordon Welchman was a pure mathematician who had had a distinguished career at Cambridge. He had left his appointment there at the outbreak of the War to go to the Government Communications H.Q. dealing with computational problems. After the end of hostilities he had gone to America to work on computers. In the spring of 1953 he wanted to return to England and approached N.R.D.C. where Hennessey rang me up and strongly recommended that we should take him on. At first he came to the Sales Department, and we had great hopes that his high academic ability and persuasive personality would help us with selling to Government and University establishments.
However, the developing enthusiasm between Ferranti management for an arrangement to make equipment jointly with Powers made it sensible that a Whitwell-Welchman partnership should be quickly established. The talks with B.T.M. were slow, partly because both groups of engineers seemed rather lukewarm and also because Ferranti engineers still had to devote a good deal of time to the Mark 1* as experience was gained in Canada and in the University and new orders were coming in. Clearly the kind of close collaboration which was being established between Welchman and Whitwell could not also be set up with B.T.M.
Meanwhile another development was steadily maturing. From the earliest days of their support of the Mark 1* there had been pressures on Ferranti by N.R.D.C., particularly Professor Blackett, and from the Company by our Cost Department to develop a "wages machine". Although the wages problem did not seem too promising to the sales staff - investigations and estimates had shown it was unlikely to be a highly profitable way of doing wages calculations, a wages machine would probably be able to do other commercial calculations. We then still thought commercial computing would prove fairly simple when we had studied it more and had the extra equipment which was needed. There was, however, the serious difficulty, for a Company with the Ferranti organisation, that the selling of small computers would mean many customers to get a reasonable turnover and these would be widely scattered. This would mean a large Sales and Service organisation which we were sure the Chairman would not agree to.
The Powers proposals however could be very convenient; they had the set-up in being.
The pressure to develop a commercial machine and his observation in the U.S.A. had led Bowden to make an arrangement with the Royal Insurance Company to study the use of computers in Insurance. A similar exercise had been begun very early on in America and the Royal now agreed that one of their most promising young actuaries, A C Baker, should be put on to the work. Tony Baker studied the problem and in March 1954 he read a paper on his proposals to the Actuary's Student Society. At a dinner given in his honour afterwards our host, Kenneth Usherwood, introduced Tony as "the only actuary in the country who was a Fellow of the Institute of Actuaries, a Fellow of the Faculty of Actuaries and also of the Chartered Insurance Institute and the only actuary who had ever made a century at the Oval". This was very useful advance publicity. The machine which came out of this study in the course - christened Perseus - was a big computer and very successful in operation but only two were made. Its coming was delayed by indecision caused by possible conflict with the Powers plans for new equipments; also it could lead to the loss of orders for the Powers PCC and it came at a time when transistors were replacing valves. Perseus was however potentially useful to the current plans because Powers had important connections with Insurance Companies and they were willing that it should be installed in some instances in preference to PCC's. A big machine of this kind was clearly needed.
In December 1952 Christopher Strachey reported the results of his experience with FERUT at Toronto and his other investigations in America. He proposed to N.R.D.C. investigations into
These proposals fitted in well with the Ferranti position: Manchester University were beginning to prepare for their next development and our own plans for using punched card experience would look after the commercial work. We had taken an order for a Mark 1* for Shell which would be used in the application of modern management methods in the oil industry and there was still a chance that the Nielson organisation would have one of our computers.
At this time the Elliott Brothers were making their 401 computer with support from N.R.D.C. It was to be shown at the Physical Society in the following spring and was to have an important effect on the future developments in Ferranti.
In the Spring and Summer of 1953 two external events occurred which changed the course of events in the Computer Group. There was a staff explosion in Elliott Bros at Boreham Wood and, secondly, Dr Bowden was designated Principal of the Manchester College of Science and Technology in September.
Elliott Bros had formed a small but powerful unit in their research department to make computers. The Research Director Dr (now Professor) J E Coales, O.B.E., who had served during the War in the Admiralty and in that capacity had designed radar sets made up of standard packages so that any part which broke down could be quickly replaced and the equipment made serviceable without attention on the spot from specialised engineers or need for workshop facilities. John Coales was now exploiting this principle and a computer (the Elliott 401) was designed and made by a team led by Mr (now Professor) W S Elliott and including a number of able engineers.
From the Autumn of 1952 there had been a danger of disintegration of the Elliott team and in December 1952 W S Elliott resigned with effect from May 1953, and Coales prepared to take up an appointment at Cambridge University. Another of the senior engineers, Hugh McGregor Ross, was recruited by Ferranti; Bowden had known Hugh when the former was in partnership with Sir Robert Watson-Watt. In due course Elliott and some of the other Elliott Bros staff went to work for N.R.D.C. and George Felton and later others came to Ferranti.
The 401 techniques could be married to Strachey's ideas. Lord Halesbury had always held that it would be easier to know how to make a small computer after building a large one. The position was now favourable. Ferranti had built large computers which were working and getting more reliable, the Coales idea had been proved in the 401 and Charles Owen, who had designed the circuitry, was steadily striving to make it more and more reliable to match the reliability which Strachey had seen in American computers. N.R.D.C. efforts to launch a small computer through Elliotts had been frustrated by the dissension in that Company.
One day John Crawley of N.R.D.C. suggested to me that we ought to take over the know-how of the 401 packages. I was surprised by this but he said that this was the property of N.R.D.C., they had a duty to exploit it and they could hardly refuse access to it to anyone who expressed a serious desire to develop the application of the patents. N.R.D.C. had given Elliott Bros "some exclusivity" in the 401 and did not respond immediately to an approach by Ferranti made in May but by September Ferranti had made an offer of employment to W S Elliott and he joined the Company about two months later. We were already well advanced with our plans for a computer centre in London which would help to promote sales as well as, we hoped, earn some money.
The Powers team did not believe a computer centre would be profitable. They had been used to giving a computation service on punched card equipment, as had B.T.M., and both were accustomed to losing money on this activity, a cost they regarded as a necessary demonstration expense. Whitwell said that "when Ferranti put a computer in London they would really begin to sell the machines". He proved right and this support came at a critical time, as some of us still believed a Computer Centre could be profitable.
Ferranti activity now fell into three main parts:
The F.P.C.1 design proceeded quickly. The redesign of the packages was fixed, the first literature prepared and by the autumn of 1954 the machine was taking shape and the first orders were taken. At the Computer Centre in 21 Portland Place we were relieved that we could now have a much smaller machine than the Mark 1* and one which ought to offer a better chance of making money.
The Sales staff in Ferranti concentrated on F.P.C.1 and preparing for the successor to Mark 1*. The Whitwell-Welchman plans would anyhow mainly supply products for the Powers Samas sales organisation and they were very secret. It was the practice in Powers to draw a blind down between the development staff and the field sales staff for the very good reason that if the salesmen knew of exciting new developments their enthusiasm for the products in production would wane. It was only when the product design was nearly final that a few special customers were sworn to secrecy and given an opportunity to express their opinion of the new products. The Ferranti Sales staff were also kept at arms length and all that was known was that a magnetic tape data processing system was being projected at a basic price of around £15,000.
In March 1954 the sales staff of the two companies made a joint report that there would be many advantages to be gained from collaboration; by September a full report on the elements for an integrated data processing system was produced for the Steering Committee of the two companies and a top level policy meeting was held. Some of the expected problems began to peep through the paper but as is customary at this early stage they were wished away. Sir Vincent told the meeting that N.R.D.C. had placed an order for 10 F.P.C.1 computers. One likely customer was the Bristol Aeroplane Company who, with Ferranti made the Bloodhound Guided Missile, and they would want to use Hollerith cards. Warwick Deacon of Powers said it would be paradoxical for Ferranti products to be linked to I.B.M. or B.T.M. equipment after a collaboration had been announced, but it had to be recognised that "from time to time" instances might arise needing to be studied "in the light of their particular circumstances". Powers pressed for an announcement of the collaboration - "the sooner the better". This was made in October and it was hoped that the differences over exclusivity would solve themselves. They did not; but if the new products had come out quickly they perhaps could have been minimised. But months and years passed without any new product showing up. Finally in 1958 as mentioned above it was to be a Pegasus computer with Powers cards which had to be put forward as the product of collaboration. This was called Pluto, to distinguish it from the Pegasus version, which was given Hollerith cards.
But the talking had to continue and there were customer problems to talk about.
To drive home the meaning of the announcement a joint exhibition was proposed. It was to be at 21 Portland Place, but the new offices were too small and the staircase not strong enough to take the heavy and bulky mechanical equipment from Powers. Another venue was proposed but there were clearly going to be separate products, not joint ones, for some time. In the end the proposal for a launching exhibition merged with plans for a joint exhibition at the Computer Exhibition in 1958.
As soon as the announcement was made Eric Grundy pointed out it was important not to put ourselves in a weak position in the United States because R H Davies was anxious to sell the F.P.C.1. He would want the Hollerith type of cards, but, in Elbourne's words, the Powers announcement meant that B.T.M. were less interested in Ferranti and we could not look to them for more co-operation than was necessary to ensure that we bought their equipment.
We therefore agreed in January 1955 that
Item (2) changed the original basis of the arrangement.
Since we could not find descriptions of the different markets which would automatically allocate potential customers to one or other of the partners we had to do this by full disclosure of our lists of prospects and collaborate as possible when the cases came up. This meant that the customers could play us off against one another to some extent. Ferranti were promoting their new Pegasus while Powers were selling P.C.C., and both were in principle intending to replace these computers with new jointly designed systems. It was a weak selling position for both companies in the market for commercial computers, but fortunately for Ferranti, Pegasus was very popular for technical problems.
There were certain immediate customer problems listed by Nash which had to be dealt withs the Ford Motor Company, the Austin Motor Company, Rolls Royce and Vickers Waybridge. Not long afterwards came one of the moat interesting, Rootes, These provided illustrations of the difficulties which arose.
Fords. The work here was in the hands of a Mr Bradley who early came to see the Manchester Computer and said he wanted to be one of the first to use this new machine. The difficulties however were very real, they wanted to do the difficult production control and commercial work for which we were not ready. Interest revived later when the smaller and more reliable computer came but we did not get their order. This ultimately went to I.B.M,(?).
The Austin Motor Company. This was to be a particularly sensitive matter. Nash had, years before, got their business, putting in their first punched card installation, Austin approached Ferranti and Swann saw them at Longbridge. They were feeling they should go electronic in due course and looked to Ferranti for help. When it had to be disclosed that Ferranti were working with Powers, Austin officials seemed none too pleased but accepted the idea of a joint study and Conway Berners-Lee and some of the Powers staff made this study but to Fred Nash's great chagrin Austins ended by buying from a competitor.
Rolls Royce, had a programmer, Griffiths, who was one of the first regular users of the Manchester Computer for their technical work, and he was very interested in Pegasus. Rolls Royce were Powers customers for their commercial work and Nash encouraged the idea of their having a Pegasus for technical calculations. They gave us a letter of intent at the beginning of 1955 and we formed a joint team including Welchman and Peter Elliott of Powers to study their commercial problems as well as the technical ones. They had a very complex spare parts problem because they had to keep spares for engines designed 40 years or more earlier.
We felt very sure of our order for Pegasus and so did Nash, but the negotiations were protracted by attempts to find a solution to the administrative problems by designing a large, special-purpose device, and were ultimately brought close by a rather peculiar effect of the organisation set up by Powers and Ferranti. To keep new products secret from salesmen was, to Powers, quite essential, and this rule was applied to Ferranti Sales Department. Their salesmen relied very much on commission for income and this meant, as Nash said, he did not really control his sales staff. The commission salesman runs a one-man business and private information is vital to earning commissions. Rolls Royce was naturally one of the companies Powers took into their confidence about new products so when the Rolls Royce Chief Administrator, Bearman, came to a meeting at which we expected him to convert Griffith's letter of intent into a firm order, he said he had just come from a meeting about equipment which he was assured would be more advanced and cheaper than the Pegasus we were offering but which he was not allowed to tell us about. Our meeting broke up and in due course Rolls Royce bought their computing equipment from I.B.M.
Vickers Weybridqe. Here Ferranti interests and contacts were at the technical level. Bowden quite early on had talked to the S.B.A.C. and they had agreed to discuss together standard forms of design calculations to be put on computers. A panel was formed of representatives of the various aircraft companies and Ferranti, represented by Hugh Rosa, were included. (When some time later it was suggested that another company should be included there were objections to having suppliers on the panel. It was pointed out that they already had Ross of Ferranti. The simple answer was given that he helped them). Although it is doubtful if many companies held to the standard forms, the work certainly helped to clarify the ways to use computers and Vickers became good customers for Ferranti computers.
Rootes. This was not one of the problems at the beginning of 1955 but it soon became one. The Rootes Group had a problem of production control and one of their staff, Hardman, had done a lot of this kind of work at the Humber plant. They were users of Powers equipment and were very dissatisfied with this. Shread, the Powers Home Sales Director feared they might lose the contract. Shread's assistant, Lightstone, was sent to look at the situation and was sharply told that the trouble lay with the equipment and not with the system. Faced with the possibility that Rootes might follow Austins Shread called for help from Ferranti. The Finance Director of Rootes was Mr Goate and he had become interested in computers. A meeting was arranged at the Powers office with Whitwell in the chair. He explained to the Rootes party the plans for new data processing equipment and it was at this meeting that he upset Welchman by his reference to the need to keep the "Ferranti designers feet on the ground". A good deal of work was done Jointly in studying the problems of the Humber Company but the combined effort to get an order broke down when Goate found he was not getting satisfaction from Powers. He telephoned me one day and said he wanted, very confidentially, to ask if we could provide a computer with Hollerith cards. I said there was no technical difficulty - we were already doing it for I.C.I, but it would of course be embarrassing to me personally. This he said was why he was now speaking so confidentially but they had decided to change over to Hollerith. I immediately wrote to Murray Robson and put him in the picture and in due course Rootes changed over to (I.B.M. or E.M.I.).
Littlewoods. The other contract which produced internal difficulties was Littlewoods. Within Ferranti Keay did an extensive study with Eddie Nash of that company (no connection with Fred Nash) and the negotiations had led through satisfactory demonstrations to a meeting to which Hedges, the Littlewood's Director in charge of the operation, brought his Managing Director to agree the order. Hedges was a former Powers man who was in charge of the O.&M. in Littlewoods and the situation seemed very promising. However, it took place shortly after a decision had been taken about the prices which should be charged and in spite of protests from the Ferranti sales staff these had been put unrealistically high; much higher than had been considered before. The Managing Director of Littlewoods said he wondered why Powers had brought him there, the meeting broke up and Littlewoods shortly afterwards placed an order with Elliott Bros and Eddie Nash later became the Sales Director of Elliott Computers.
This was a particular disappointment for several reasons. The installation was to have been in Hollinwood which would have been very convenient for the development of new equipment; the market for computers in chain stores was clearly going to be a big one, and it would have helped to consolidate the Ferranti-Powers arrangement. The profit margins asked for by Powers were necessary if they were to continue against our advice, their practice of doing all the programming for the customer. This was the way the punched card firms sold equipment, and was probably necessary in dealing with the majority of their customers, but we had strongly urged that it was too costly for computer salesmen to study each potential customers' business in the detail demanded; nor was it desirable. The customer would have to learn to program the computer themselves if they were to get the most out of the machine and the economic procedure was for supplier and customer to work together, the latter supplying the detailed knowledge of the business and learning to write computer programmes under the guidance of an expert on the computer. Powers, however, considered their way was professional selling. They seemed to regard our way as showing weakness.
There was mounting dissatisfaction with these dealings in collaboration from which no single order resulted and no doubt there was much unhappiness on each side. It is doubtful whether most punched card men thought computers more than a flash in the pan at this time.
There was one order - for a Perseus - which slightly relieved the gloom. Powers had an old customer in Cape Town, the South African Mutual Life Assurance Society, who had sent one of their young actuaries, Peter Bieber, on a world tour to study computers. Passing through London on the return journey from America Bieber was told about the F.P.C.3 insurance computer designed by Ferranti.
By a coincidence, Hugh Ross and I were shortly to go to South Africa. Grundy had made an engagement to visit Johannesburg to talk to their Institution of Electrical Engineers but pressure of business was making the journey very difficult. To avoid disappointing Professor Bozzoli of Witwatersrand University it was suggested that I should go and deliver the lectures, aided by Hugh Ross. We were due to go in October 1955 and Peter Bieber made us promise to visit SAMLAS. We were able to arrange this and found Mr Murray, the General Manager, very receptive. Bieber, he said, had sent back reports that they should spend £250,000 on a Univac computer but Lloyd, their actuary, had now learned that we could provide better value for money. The upshot was that we received a letter which said that if we could make the computer described to them for £200,000, they would give us the order. We had not gone to South Africa and Rhodesia to make contracts, we were not really ready and the proposal came as something of a shock. It now became a question of whether we should sell Perseus, or two or more of the tape processing systems which were being projected. It was about ten months later and after a good deal of urging from SAMLAS that we accepted their order and manufacture could go ahead. Lloyd later, and rather hurt, said it had taken him 20 minutes to persuade his Board to order our computer and 10 months for us to accept the order. Now that we know the costliness of both making and of buying computers, the Ferranti attitude becomes more easily acceptable, but it did not help sales. Another order from Sweden was taken, but the long delay in getting agreement as to whether to supply Perseus or not had now killed the prospects for the computer. Valve machines were on the way out. The two Perseus computers were, however, technically very successful and if made them quickly would probably have sold well during 1956/57. When we began talks with B.T.M. again in 1959. they were preparing to sell a large valve machine and Perseus would have been available earlier and probably have been a better machine for large scale data processing in insurance and other file processing work.
It is interesting to look at the history of the joint efforts from the point of view of a natural desire on Powers part that Ferranti should use only Powers input/output equipment. In September 1954 Deacon had declared and Sir Vincent had accepted in principal that it would be paradox!a1 for the two companies to announce a collaboration and for Ferranti products to be linked to I.B.M. or B.T.M. equipment, but there were customers in the U.K. and more particularly in the U.S.A. whose needs might make it necessary to modify the policy. By January 1955 the rules were altered to say that
The "at least" was a recognition that if work had been put into launching equipment using particular ancillary devices it could be very expensive to withdraw the resulting products. It was also recognised that while there was an understandable difference between scientific and commercial work, it was impossible to implement a selling arrangement based on dividing customers on a dictionary definition, which was too academic to interest them. Customers were more concerned with which supplier could, on technical and commercial grounds, best meet their needs than on the way Ferranti and Powers made arrangements between themselves. Since most computer customers had both kinds of work the distinction was an invitation to play off one company against the other.
A big meeting to work out these various marketing problems was held in April 1955 at Straford-on-Avon. Here preliminary specifications of the joint product were tabled, and arrangements made for Shread, and Swann to compare their lists of prospects. It was suggested that the Powers organisation was the logical outlet overseas and that consideration should be given to the supply of suitable equipment for Ferranti Cost Department. When this last was interpreted as to supply Powers P.C.C. computers there were objections from the Cost Department, which used Hollerith cards, and from the Sales Department which could not face telling potential customers that Pegasus was not suitable for Ferranti. (It was at one time argued that our market was Universities and Government Research Establishments only). The overseas selling problem seemed simpler and lists of agents were exchanged and letters issued setting out how the two companies would do business on similar terms. (There was a little local difficulty in Italy where the Powers agents, Logomarsino, were pushing Elliott computers, but it was hoped this could be wished away).
The designers would not accept, as was being accepted by the Ferranti Pegasus team, that existing equipment, particularly magnetic tape drives, could be bought from the United States. Pollard had earlier objected to the offer from N.R.D.C. that they should give us American equipment to attach to the Mark 1* computer and was now followed by Gutteridge, at Powers who said they must have a design which was "peculiar to and precisely in accordance with their own exact needs. It was not feasible to take an equipment from some other manufacturer's stock". There seems to have been no explanation given of this statement.
By the autumn of 1955 Pollard was becoming concerned at the extent to which it was being said that Powers had the exclusive agency for Ferranti and wrote that "so far as computer products were concerned we must maintain complete independence and freedom to market in any way we chose" . A year later marketing policy was modified; if a customer wanted a Pegasus computer to work with a system other than Power Samas, we should be free to supply and equally we should not object if Powers Samas wanted to sell this equipment with any other computer. Where a customer had no punched cards we should push the joint products. While Ferranti agreed not to work with I.B.M. exclusively was otherwise virtually forgotten.
However, no orders came from the joint activity which Powers claimed was considerable and expensive, though by March 1957 Pollard was writing that the "Powers efforts were more than pathetic. They did not seem to have made any serious attempt to build up a selling force". A modification of policy was agreed between Nash and Prytz (Powers Export Director) for Powers and Robson and Swann for Ferranti:
Meanwhile in a rather Canute-like phrase, the companies were to pursue a policy of "persuading the market to wait for the joint products, wherever possible".
October came and a decision to continue with designs of four Tape Processing Systems and that sales of Pegasus with Electro-Data tapes should not be stimulated. Prytz was clearly losing interest and his response to an item on the minutes was that he "would find the papers when he had time". Swann wrote to Robson "It is impossible to guess what Powers feel. I suspect there are nearly as many different views as there are people at the senior level". Welohman returned to America, and there was a meeting of the Steering Committee of the two companies at which Pollard said that if anyone had to make sacrifices it should be Ferranti. Robson's response to this effort to pursue the partnership was to decide afterwards that he and Swann had better deal with future meetings.
Throughout the negotiations the question of prices was a problem. Ferranti claimed they could not give more than 15 per cent discount to Powers, the same figure we had agreed with Ericsson who represented us in Sweden, but Powers wanted a margin of at least 25 per cent on the selling price.
On the assumption that they would, as they did in the case of punched card equipment, do all the programming work, this was not unreasonable, but Ferranti considered this a wrong policy. Also since Powers salesmen worked on commission, they could only spend the long time needed on building up a computer clientele if the rewards were substantial. Commission on other equipment could go up to 36 per cent or 38 per cent. This was much too high for a product selling at £40,000 - £45,000, but in the Sales Department we had always agreed that the selling price should be twice the ex-factory cost to leave a margin which could if necessary allow up to 20 per cent for selling.
When the P.P.C.1 was first projected the costs were estimated at about £16,000 and it was thought that a selling price of twice this figure would be sensible because it would lie between the Elliot 401 and the English Electric DEUCE. Lord Halesbury had wanted to be able to sell the machine for under £30,000 but he saw this would be impossible and the first figure fixed, though not the first formally quoted was £32,000, including power supplies and the necessary input/output equipment.
When it became necessary to fix a price for Pluto, the standard Pegasus without power supplies had risen to £42,000 which meant that Ferranti could sell to Powers at £36,100. Since the latter wanted a common selling price to yield 25 per cent on the selling price this had to be over £48,000. If the same formula was applied to a basic Pegasus computer this would have to sell at £6,000 or more than the English Electric Deuce.
These prices compelled the setting up of a small sub-committee (Lightstone from Powers and Keay from Ferranti) who reported in June 1958 that the suggested prices would have to be reduced by £40,000 to sell for £140,000 or £145,000 with punched cards added.
By now these figures would only have been of interest if they had promised a good profit. Powers had decided to join with B.T.M.
When in 1955 Lord Halsbury proposed the organisation of a Computer Exhibition, this was to be more than a demonstration of a new phenomenon in engineering. It would proclaim the beginning of a new industry. The formation of a Computer Division within the Electronic Engineering Association was a recognition of this.
The first Chairman of the Computer Division was Clifford Metcalf, a Joint Managing Director of E.M.I., and the companies making computers or parts were members; in particular Ferranti, English Electric, Elliott Bros, S.T.C. I.B.M. joined later.
For the Exhibition to cover the whole computer business it was necessary to include the punched card firms and the manufacturers of ancillary equipment associated with computers. This brought in the Business Equipment Trades Association (B.E.T.A.) and B.E.A.M.A. A joint Steering Committee was set up with two members from each of the three Associations, Metcalf and Swann representing the E.E.A. Two other Committees were formed to deal with the organisation of the Exhibition and with publicity.
Planning for the Exhibition showed clearly how interdependent the new industry was. There was a lot to be gained by fighting a common enemy and common enemy was the ignorance and consequent reluctance of the potential customer. If computers excited technicians, there was a good deal of nervousness about the money they would cost and uncertainty about the returns they would yield. The Exhibition was another milestone. Powers and Ferranti occupied a large stand jointly and Powers put on display relating to the Pluto computer, which was a Pegasus with Powers cards, and had originally been intended to celebrate the computer marriage of the two companies. By then, however, pressures of business developments and, I think, the lessons learned in planning the Exhibition had shown the two punched card companies the folly of continued competition to the advantage of I.B.M. and the computer manufacturers; and B.T.M. snatched the Powers bride away during the celebrations.
The Pluto became Ferranti's baby and after some effort Harry Johnson sold it to the London and Manchester Assurance who found that the card equipment gave a good deal of trouble and that the design lacked the full facilities of a Pegasus 2. They christened it "Pegasus 1.7/8" and later sold the Pegasus unit to Newcastle on Tyne Technical College, without the punched card equipment.
To go back to about 1955, there was also formed, on the initiative of R H Williams of Computer Consultants, a small luncheon club called the "Nought One Club" which met occasionally. The representative of English Electric on the E.E.A. and at the Nought One meetings was E R Davies. He had once or twice talked to me of the importance of co-operation in the industry and then one day in September 1957 after a Nought One meeting attended by the head of the Treasury Computer Support Group, Davies asked me if I would consider joining English Electric and would I call at Stafford to meet the General Manager of the Stafford factories. I said I would be happy to meet his General Manager as I hadn't been to their Stafford offices since I went with Bowden in 1951, but I did not think I would consider changing employers. At Stafford it was clear they were bothered about the prospects because the English companies were all competing for both customers and for staff, depressing prices and increasing costs. I was told how they had tried to get an arrangement with B.T.M. and had failed, though their computer was linked to the Hollerith cards. I explained that, as was well known, we had an arrangement with Powers but it was clear that E.E. knew this had been unproductive and probably from B.T.M. that it was unlikely to get anywhere (B.T.M. were by this time getting together with Powers).
Shortly afterwards, during a visit to E.M.I, in connection with the Exhibition and to see the equipment they were preparing under Robin Addle, Clifford Metcalf asked me if I would join E.M.I. I replied here also that I would like to see more co-operation and less competition in the industry which was producing too many models of computers, and I suggested we might arrange an exercise in joint marketing in Australia which was a long way from our other markets but would become important very soon. The Chairman of E.M.I., Sir Joseph Lockwood, liked this idea because they already market records in Australia for Decca who were their keenest rivals elsewhere. We had some discussions between the three companies and it was agreed that we would try and arrange conversations at the top. After Pollard's resignation Metcalf and two of his staff entertained Peter Hall and me, and Metcalf again privately asked me to join E.M.I, and repeated the request sometime later when we were all at an E.C.M.A. conference in Brussels.
I realise I should have accepted this offer at the end of 1958. In fact this was probably the best point in time at which I might have got agreement to an arrangement which I was sure the industry needed. Had we made it I.C.T. would have had to join in and a powerful British computer company would have been formed eighty years earlier than was achieved by the creation of I.C.L. after millions of pounds had been spent on proliferating models of machines which the industry had inadequate power to produce fast enough to keep pace with technical developments. While the British companies weakened one another, I.B.M. strength grew. We had held off their first attack on our market with the 650 because Pegasus was a better computer at about the same price, but I.B.M. produced the 1400 series and began to install these machines from about 1960 on. Customers could go to I.B.M. for a range of machines nearly equivalent to what the whole British industry could offer. The 1400 was probably partly accident. When I asked Kenneth Barge, I.B.M.'s Sales Manager, about it he said he did not regard it as a computer but rather as a card-to-tape converter. Users found that the facilities needed to do the conversion produced a very useful computer.
Ferranti management were always convinced that our problems lay with selling. This was why they wanted to link with Powers and now Peter Hall opened conversations with I.C.T. In Sales we were less enthusiastic and I did not enjoy the prospect of more years of being told not to sell our products. To me the problem lay with production, we were always late with deliveries - even by 1963 we were not able to deliver a Pegasus to the Westminster Bank on time!
Useful work was done in comparing the proposed Orion with a computer I.C.L. were proposing but even in this something of the old pattern of procedure showed. Ferranti showed the I.C.T. engineers everything about Orion, but our staff experienced difficulty in getting information about their computer. One day on a visit to Edinburgh I met Sir Cecil Weir, Chairman of I.C.T., after breakfast in the hotel. He asked me how the talks were progressing. I told him we had given his people all our information; he asked if his staff were being slow and said he must speak to Cecil Mead. He clearly was fearing some obstruction and the way was cleared for the joint study to be speeded up.
When Sir Edward Playfair took over the Chairmanship of I.C.T. in 1961 he proposed at a meeting at Manchester that our companies should agree to work together by dividing the market. I was unenthusiastic because the proposal was to give Ferranti the large computers and I.C.T. the market for small ones, the market for medium sized computers depending on rather vague goodwill between the parties. This could only mean that once again Ferranti would be expected to give way whenever there was a problem, because Ferranti "did not have shareholders who could make objections".
I expressed my view when Sir Vincent telephoned me about the arrangement. He said all he had got out of the Powers deal was a fancy nutcracker for Christmas (I had one too) and I said that the vague "working together" arrangement could not work when our two teams of salesman faced the common enemy in the form of a hesitant potential customer. The only sensible solution I could see would be for I.C.T. to buy the Ferranti business if they wanted it.
The possibility of an arrangement between the engineering companies did not die. At the Computer Exhibition in 1961 English Electric's Manager of Stafford factories saw me and invited me to call again to see him at Stafford. Before I could do so however he died in a sailing accident at Bala. E.M.I. still wanted an arrangement with Ferranti. One day I met Sir Joseph Lockwood in the train coming from Manchester. He had been talking to Grundy and it seemed that Lockwood was hopeful. He has since told me that conversations continued between E.M.I., Ferranti and I.C.T., but these failed because Ferranti were unable as a privately owned company to join with two public companies.
E.M.I, now joined I.C.T.
Christopher Strachey had become interested in computers while teaching mathematics at Harrow, when he heard about and came to work on the Manchester Computer. He joined N.R.D.C. to study computer design and was lent to the University of Toronto to work on the Mark 1*. This experience was reinforced by a study of the computer scene in the United states. When he came back to the U.K. Elliott Bros were working on what was to become the Elliott 401.
This computer began as a test bed for the ideas of John Coales who, was developing his techniques of breaking down the circuitry into separate packages which could be plugged together to make the final assembly. The repair problem was thus changed to finding the package and replacing it from a stock of spares.
This two-stage fault-finding has two big advantages:-
Since the first stage only involves finding the package which is faulty, this can be replaced by a spare and the work can continue. Getting the machine "on air" again is of the greatest importance in computation work, and the faulty package can be repaired at leisure and, if necessary, by sending it away to a fully equipped workshop.
The package technique had been criticised on a number of grounds:
It was suggested that varying delays and amplitudes in pulses generated would call for a lot of adjustments; that the redundancy in the computer would be very wasteful of hardware; that the plugs and sockets would cause trouble; that the separate test equipment would be costly.
These possibilities had to be allowed for but W S Elliott, who was in charge of the computer was careful to choose pulse frequencies well below the limits for the circuits and the delay lines; the extra packages which were a measure of the redundancy amounted to only about 10 per cent and there were in the original machine only 5 types of packages and these numbered 150 in all.
Dr Bennett examined the 401 and pointed out that one of the great advantages of the technique was that special machines could quickly be designed by using this building block technique and this redesign could be done at a distance from the works. This could make it easier to make new designs in close collaboration with the customer. Although we were not thinking of a large number of "specials", the same argument applies and became important when various peripherals had to be connected. It followed also that the designing of computers could be done by logical arrangements of the "building bricks" so the engineers were not needed at all stages but could concentrate on improving and developing the techniques.
After the first demonstration of the "401 techniques" the computer was sent to Cambridge where Strachey did a considerable amount of redesign work. There was a suggestion by Grundy that Wilkes the head of the Computer Lab there had little confidence in this work so in August 1953 I wrote to Maurice Wilkes who replied very promptly - from Cambridge Massachusetts - saying that he was satisfied that in the hands of Elliott the electronic design had be shown to be sound, but he was waiting to see what happened when Strachey's wholesale redesign of the logic was tested. As Strachey was not an engineer, if the packages did not then give trouble it would show that "the design of the basic units was of a very high order indeed". The redesign was successful. This was demonstrated when the first model of Pegasus was made without any prototype having been required and continued in use for thirteen years.
Before writing to Wilkes I had written to Grundy saying that I thought a machine of this type would be desirable. I had had an approach from the Government of India statistical adviser, Professor P Mahalanobis F.R.S., who I had met in Thailand a year earlier and thought that a 401 type machine might do his work and a lot of other jobs, including wages. I therefore proposed we should formally ask to have the 401 techniques and also if possible get Elliott himself.
Pollard was not pleased that Grundy decided to recruit Elliott, who, in turn, was not pleased to have to work under Pollard. (It was rumoured, though I have never checked on this, that at one time Pollard had applied for a job under Elliott, who had turned him down). They were both individualists, there was little sign of personal sympathy between them and neither wanted to take account of the other. Grundy and the Sales Department wanted the abilities of both because the products they could make were. from a marketing point of view, complementary. Although the Manchester Univeristy Mark II seemed then to have a limited market, the success - for those days - of the Mark 1* showed that this view could be wrong. The two engineers could perhaps have worked satisfactorily if each had been independent and Grundy tried to achieve this by agreeing that Elliott should have his own unit in London. But he was subordinate to Pollard and his unit was only a development staff and Elliott wanted a production unit was well. Pegasus was designed for production and the first one - a production model - which was built during 1954-56 and was to continue in service until the summer of 1969. Elliott was right but perhaps for the wrong reasons; he was very conscious of the limited power the commander of a small unit would have. The real benefit from his plans were: the sales advantages of being able to introduce the development and production staff to potential customers; the more speedy solution to problems if the two were close together and the advantages of having an extra source of supply. The main production could have been concentrated in Manchester when all the problems were solved. In practice there were many delays caused by setting up a production line too quickly and for largely political reasons. Pollard argued that one of the benefits to be gained from a packaged computer would be the need to make a large number of similar packages and these would be more cheaply produced in Manchester. But Elliott was, I believe, sure there was a political motive; the production unit would be large and power would lie where the weight of investment and numbers of people were found. The trouble only showed later but the consciousness of it influenced relations from the beginning.
Elliott was justified by the subsequent history. The production line took a long time to get going and costs were so uncertain that for a period of some 9 months in the middle of the sales drive we were forbidden to send any quotations. The selling of Pegasus I never properly got going again and the loss due to the disruption was very heavy.
The Sales Staff wanted a saleable product and Elliott seemed to be offering it. The Manchester Mark II, which became the Ferranti Mercury, was a more doubtful prospect. There was a small market for a power computer but selling a small number of very expensive units is difficult and seemed dangerously so in 1953; in the event we got away with Mercury because of the demand which arose for computers in the field of nuclear energy. Relationships between Sales and the Elliott's organisation were also easier on two other counts: former employees of Elliott Borthers came to both Engineering and Sales when it had been agreed that a sales office and computing service should be established in London. In Manchester there was always something of an "anti-sales" attitude: Manchester University always felt that so remarkable a product as a computer should sell itself, and were impatient of "sales points", if they interrupted the research work. The F.P.C.1 staff were only too anxious to have their product sold and eagerly welcomed the offer made by N.R.D.C. to buy 10 machines, while Pollard rejected their advances over Mercury; he persuaded Sir Vincent he would make so much money out of it that it would be folly to share the profits. For some reason he was convinced Mercury would be easy to manufacture.
Following his success in redesigning the 401 Strachey now persuaded N.R.D.C. that a more ambitious machine could be developed with Ferranti. This should be designed with certain definite objectives in mind. 1) There should be no doubt as to whether a fault lay in the hardware or in the program, 2) optimum programming (having to arrange in which address in the store each instruction had to be placed to ensure minimum loss of time) was to be avoided because it tended to become a time-wasting intellectual hobby of programmers and 3) the needs of the programmer were to be a governing factor in selecting the order code. Finally, 4) the computer was to be cheap.
In 1953 reliable core storage was not developed. The storage methods in use were the c.r.t. system in Manchester, mercury delay lines on EDSAC at Cambridge and the N.P.L. and the nickel delay lines used by Elliott's in the 401. The last were now shown to be satisfactory and suited the proposed design because they could be accommodated in packages of the same size as the circuit packages. Magnetic drums were available to provide the necessary large backing store.
Fast stores are costly so money could be saved by minimising the size of the fast store and facilitating the transfers between this and the drum backing store. The Mark 1* transferred in blocks of 512 words, which presented a problem as to which block of information to keep in store and where to put them. On Pegasus it was decided to make the blocks small - only eight words and that six of these blocks should suffice for the bulk of the computing store. (Originally Strachay had planned to have only four eight word blocks but George Felton demonstrated that six would give a much greater power). These blocks could be transferred between anywhere on the drum store and the computing store.
There were eight accumulator registers. Accumulator 0 was special, its content was always zero and it was sometimes called the dummy accumulator because it was not a delay line. Each of the other seven could be used as an index register or B-register and the arrangement of the modifiers enabled the programmer to handle transfers of data in and out of the small immediate access store with little more effort than if he had a large single level store. The type of order code devised to suit this arrangement was repeated on Ferranti Argus, Orion and F.P.6000 computers and hence on the I.C.L. 1900 series.
The small computing store was adequate because the computer was given a large repertoire of instructions which reduced the length of programmes. To ensure accuracy the idea of a parity check was introduced; this was kept on each word in the fast store and on the drum. This new development greatly helped to determine when a fault was due to the hardware or to the program and consequently reduced the tensions between engineers and programmers.
Another facility which helped to ease the programmers fault finding was the use of an "optional stop" facility. The Pegasus word was 39 bits long, so that numbers were 38 bits plus sign and each word could hold two 19-bit long instructions and a spare bit. This spare bit was used to mark points at which the programmer wanted to halt the machine so that he could operate it order by order from one stop to the next.
The two accumulators 6 and 7 had special facilities for dealing with the double length numbers resulting from multiplication. There were also a set of registers for special purposes relating to the hand switches, the input/output equipment and to hold parameters which would be very frequently required and which therefore would otherwise have had to be frequently stored, increasing the instructions and consuming space in the store.
The "ease of programming" reputation gained by Pegasus was quickly reflected in the attitude of the users and this was not only in Ferranti, where George Felton led a team who gained a reputation for skill and enthusiasm. It is a common comment on Pegasus that users not only were able to do a lot of work but they felt a strong affection for the machine.
Looking back at the names in diaries at dates before the F.P.C.1 was contemplated and indeed while the team which was to make it was still engaged in finishing off the Elliott 401 shows how useful the demonstrations of the Mark 1* had been. Among the organisations with which we had contact and to which we sold computers were:
Negotiations with these illustrate the market situation at the time. At the Admiralty, as soon as Dr (now Professor) Vajda was given the specification of Pegasus he arranged to buy one. The R.A.E. was interesting as illustrating how a sale could be helped by efficient support from the engineers. C A Wass, who was in charge of the mathematical section, telephoned me and said that they had virtually decided to buy an Elliott Computer but as a good civil servant he felt he should check on what we were offering. We arranged to receive him at Manchester Street and Elliott organised a team, nominating one of each group designing the parts of the computer - circuitry, magnetic drum, magnetic tape system, input/outout equipment, while Felton from the Sales Department demonstrated the programming facilities, so all the questions could be readily and fortunately satisfactorily answered. The R.A.E. team left satisfied and placed an order with us. Short Bros showed an early interest in using the Manchester computer for technical computation and some of the work done for them gave dissatisfaction. However, they continued to give us service work and when Peter Hunt joined our team to work on Pegasus early in 1954 a very promising relationship developed and a good deal of work was done for the Belfast works. Peter Hunt worked hard on technical problems with Foody of Shorts and when I went to see the Shorts Board in Belfast negotiations to sell them a Pegasus apparently went well. Then one day the Technical Director, Keith Lucas, came to my office. His reputation then stood very high; he had recently designed the "Flying Bedstead" which was the precursor of the vertical take-off aeroplane. He told me that he had come to ask a few questions after which he was to go to see English Electric to appraise their Deuce computer and the Board would accept the choice he made. Unfortunately, Shorts had an important contract from English Electric. Later we were told that the Board, acting on Keith Lucas's advice, had decided to buy a Pegasus and instructed their Chairman, Admiral Slattery, to tell Sir George Nelson, Chairman of English Electric, of their decision. The result of the meeting of the two Chairmen was that Short Bros ordered a Deuce. Admiral Slattery was the one member of the Board of Shorts I had not met on my visits, but several years later at a luncheon at Olympia we were introduced by Sir George Mallaby, the former secretary of the University Grants Committee. Admiral Slattery said he had thought I would not speak to him. He was then Chairman of B.O.A.C. and if this meeting had been earlier I could perhaps succeeded in helping Ferranti Packard to sell a seat reservation system to the B.O.A.C.
The I.C.I, contact was with Cyril Wilkes at Blackley. It developed until in due course he wanted a data processing system for the Dyestuffs Division. This involved making a punched card to magnetic tape converter which also incorporated a 150 line a minute printer made by Bull in France. This was the precursor to the Pegasus 2 data processing system, and for some time it gave a fair amount of trouble and Wilkes, though a good friend, was a critical one and perhaps expected a rather unreasonable standard for the times. It was in these early systems difficult to sort out how much of the meeting troubles was due to the electronics, how much to the punched card equipment and how much to the separation of the manufacturing of the system from the designers.
The Northampton Polytechnic computer was provided by N.R.D.C. because Dr Bridges had pioneered the study and teaching of programming in London and had put on some of the earliest courses and presentations of new computers. A very close association grew up between the Polytechnic and the industry, including the British Computer Society.
As Chairman of the Committee which bore his name, Sir David Brunt was one of the most important influences in the Universities. Sir David was very pleased with the facilities which Pegasus promised, particularly the aim of making it easy to use. He felt strongly that students' time should not be spent on ingenious ways of coping with the vagaries of unreliable machines and for the Universities which could not justify a big computer like Mercury, he generally recommended Pegasus. He used to tease me that it was he who sold computers to Universities, not Ferranti, and there was a lot of justification for his claim.
However, all these orders were to come later. In the middle of 1954 the proposals had not been made final when Dr John Allen Ovenstone came from Australia to see us. He was a graduate in medicine - from the University of Adelaide I think - who had become interested in computing and was now wanting a machine for the Long Range Weapons Establishment at Woomera. He liked the ideas in Pegasus but alleged that it was unduly complicated and wanted us to make a simpler, one-address computer. Halsbury would not however be diverted from Strachey's concepts and Ovenstone bought an Elliott computer. Our failure to get this order was to have repercussions some years later. Immediately, too, it produced some worry. When the matter was discussed at a meeting, presided over by Pollard and attended by John Crawley from N.R.D.C., Owen was cautious, emphasising the complexity of the circuitry and the consequent difficulty of identifying faults though the package system should make an identifed fault relatively easy to cure. I suggested that there could be sense in proceeding with both designs because the early days of the Mark 1* had made us fear very much the effects of any unreliability, I was sure we would never be able to stay in the business if this happened again. Halsbury was, however, fully confident of Strachey's work and determined to support it and the contract was given to Ferranti on this basis.
By the autumn of 1954 the design was settled and N.R.D.C. had given us an order to make 10 of the model.
From the time when N.R.D.C. gave the contract to the day when the F.P.C.1 did its first calculations was just about 12 months, although it was over six months later that the magnetic drum was working and the computing service could begin.
The basic speed of elementary operations in Pegasus take:
Addition and Subtraction 0.315 ms Multiplication about 2 ms Division about 5½ ms
Pegasus proved popular and 38 were made. One half were sold for research work in Government and industrial establishments while seven were sold for aircraft calculations akin to research. Only 4 were sold for commercial work, but we were unable to seek orders in this area because of the Powers agreement and we could negotiate only with customers who approached us. This probably partly explains why customers and others have often complained that Ferranti selling failed to exploit the advantage they were given in the early fifties; it is commonly said that their computers were not sold but bought. There were other reasons: the business could not be used to make the more moderately priced and easier-to-maintain computers which the times required; sales were restricted by difficulty in meeting delivery promises; uncertainty as to costs led to our being forbidden to quote during a period of many months after selling had started very well. Our policy of "selling from the shop" in Portland Place rather than having staff touring the country and selling on customer's premises, a policy we adopted deliberately because it was the cheapest way of keeping contact with large numbers of potential customers, may have added to the reputation. We regularly had 50 - 80 visitors a week in Portland Place which would have required a large staff it we had to visit customers at their premises. When Powers ended the arrangement just after the 1958 exhibition, for which we had prepared joint selling arrangements with them, we were faced with dealing with commercial work for which we had been unable to make any proper preparations.
It remains true, however, that we ought to have been able with a satisfactory production and sales organisation to sell twice as many Pegasus computers and if we had then transistorised this computer, very many more.
A large number of visitors to Portland Place came to learn programming and to try out solutions to their problems. Many continued as long-term users. If they decided to buy a Pegasus they wrote programmes and came to try them out. They could thus often start up their computer work before they had delivery of a machine and if delivery was delayed we could offer machine time on special terms in compensation. Some customers continued to use the service after they had their own computer because this became so quickly fully occupied. These various categories particularly customers awaiting delivery, became important contributors to the computing service income and were compensation to the service staff for the help they had to give the salesmen.
The phrase most commonly used in relation to Pegasus is the affection it inspired in users. It was a work horse with little or no ill-temper. Also, its size seemed to suit a large number of programmes. The result was that these computers continued in service for long periods. The original Pegasus, designed and built in 1954 - 1956 came into full operation in June of the latter year. When I.C.T. took over it was sold to Vickers Armstrongs as an extra machine and they in due time presented it to their neighbour, the Brooklands College of Technology with whom they had close relations, and it continued to work there until 1969 when Vickers offered the College the more powerful Pegasus 2, itself then eight years old and which is still at work. All electronic computers can be kept going virtually indefinitely if spares are available but the early ones were pensioned off because they occupied valuable space and maintenance became too expensive. Pegasus computers continued longer than most because it was not too expensive to maintain and also because buyers could have access to the large group of programmes available with it. A sufficient library of programmes enhances the value of a computer and may, as it certainly seemed to do with Pegasus for a time, more than offset the fall in value due to obsolescence. It helped to lengthen its life.
It took patient effort in the early days to persuade customers that a computer would have a practical life even as long as five years. Accustomed to think in punched card terms and contracts which enabled them to replace tabulators etc as new models came along, they often wanted to rent computers. Ferranti financial policy demanded that we should sell them outright and we could, with truth, show that having incurred the expense of writing programmes and implementing procedures, they would want to keep these with only essential modification for a number of years; five seemed a good estimate of the minimum. This could be accepted but of course renting had other advantages: 1) the main computer might remain, but renting improved ancilliaries could have advantages and 2) accountants were often readier to pay rent than incur capital expenditure.
It was with Pegasus particularly and later with Mercury that we began to appreciate that computers could have long and valuable lives and a number of attempts were made to organise rental terms which would be acceptable to Ferranti and the customers. These would have brought the full selling price plus interest involved in 4 years or less and the rest of the rental period which would on average have continued at least as long again which would have produced a valuable extra cash flow. In the 1950's however the risk of providing the necessary working capital was too great.
Although Perseus was not started until Mercury was available for sale, it is convenient to insert a note on it at this point.
As described above, Perseus began with the study Tony Baker made along with John Bennett of the kind of computer needed in insurance, By the time it was decided to make the computer Tony had returned to his job at the Royal Assurance and John Bennett had accepted a lectureship at Sydney University where he subsequently became Professor of Computing Science. Perseus was therefore designed by Hugh Devonald and Peter Hunt with the advice of Harry Johnson. The two models were made at the Lily Hill laboratories at Bracknell.
The well-tried packages designed for Pegasus were used, but Perseus was much bigger than Pegasus since it had to store and handle what was at that time considered to be vast quantities of data. The increase in size was a considerable test of the packages, which was successfully passed.
Although Perseus manufacture was initiated by the order from South Africa and supported by an order from another insurance company in Sweden, the design was intended to make it suitable for a variety of commercial jobs which involved processing large quantities of data.
Each character - letter, decimal digit or other symbol, requires six binary digits and in Perseus these were grouped at twelve 6-bit characters for data or as three 24-bit orders making a word-length of 72 binary digits.
Conversion to and from binary was avoided because it can be so wasteful of time. Perseus worked directly with letters and numbers which were held in the machine just as they were in the office outside. Also it could work in any radix specified and also perform mixed radix arithmetic. In commercial work it is necessary to be able to use pounds and fractions of pence (when Perseus was designed U.K. currency was pounds, shillings and pence and complicated currencies have to be allowed for) as well as, e.g., hours, minutes and seconds. Also Perseus design permitted parts of words to be dealt with so that time was not wasted in processing those parts which were unchanged.
The order code was based on the wide experience gained by Ferranti programmers particularly on Pegasus so that a Perseus order closely resembled one for Pegasus. Optimum programming was made unnecessary as with Pegasus. There was a range of 63 order.
Addition, subtraction and most of the organisation operations took 234 microseconds and both multiplication and division were autonomous so that they were carried out simultaneously with other operations, increasing the effective speed. Also Perseus could operate on parts of words so that information could be packed with more than one item to the words. This field selection property was very important in commercial work.
The store was in two parts. The computing store was constructed of wire delay lines and the main store held on magnetic tape. The former had a capacity of 1024 words arranged in 32 blocks of 32 words each which appeared to the programmer as a single store. In fact, however, it was in three parts:
The main store consisted of 16 magnetic tape mechanisms. These were grouped in fours, each group being controlled by one tape control unit with its own buffer store. A 1/2 in magnetic tape was used, in lengths which were multiples of 600 feet up to a maximum of 3000 feet. The maximum length held 2,380,000 alpha numerical characters or 24,800 card images. The information was in 32 word blocks, each separately addressed and each reel had an identification space used to ensure that an incorrect reel was never used.
Input from 5- or 7-hole paper tape and punched cards. In practice the two models made used Powers cards. It was found that, to ensure continuous running of the computer, a higher standard of efficiency was needed than was required for satisfactory punched-card working; another example of the engineering problems which arose when well-tried equipments were brought in to work with computers. Perhaps it was this kind of problem which had led Gutteridge with his punched card experience to say that computer ancillaries had to be in accordance with the computers' exact needs.
Output was to a teleprinter or by means of the high-speed Powers Samastronic printer. This was an ambitious development which took time to get working but proved very reliable. We felt considerably nervousness when in Sweden the printing for the two insurance companies was dependent on a single printer, but it worked.
The need, in commercial work, to check each operation was accepted. All arithmetical operations were checked; a parity check was put on each word in the store; punched cards were read twice and compared; and a self-checking paper tape code was used. On the magnetic tape check sums were formed and compared for each block. After writing on tape information was immediately read back and checked automatically. If an error was detected the order which led to it was repeated, if necessary several times, and if the error was still shown to be present the computer stopped.
In operation Perseus was very successful but the Ferranti Powers research work had been concentrating on the idea of making a small, cheap, but powerful tape processing system which did not materialise and the opportunity to promote this powerful Perseus system was lost. It gave in the end a clear demonstration of the effects of divided loyalties. Ferranti would have liked to sell the system, and the insurance industry was preparing to accept big expenditure on computers, but the obligation to Powers left the initiative in their hands and they showed no interest in selling Perseus rather than their own computers, but presumably would have welcomed more orders for the punched card and printing equipments. All attempts to exploit the complementary interests of the two companies came up against the wall of secrecy which surrounded Powers selling policy and practices.
The designers and users of the Mark 1 computer and of its near-copy, the Ferranti Mark 1*, learned a lot in making and operating these machines. The University programmers under the energetic leadership of Dr Turing and R A Brooker developed new programming techniques while the engineers slowly identified the problems of making thy complex equipment more reliable. This involved an enormous amount of hard work for Pollard and his team, particularly Lonsdale and Hodgkinson. Many of the troubles in Mark 1 were eliminated in the Mark 1* and the average time between faults grew longer. Whereas in the early days programmers could only be sure of good spells of a few minutes these lengthened to about half an hour.
Kilburn and his team were now - about the end of 1952 - ready to take the next step and design a computer with certain improvements which appeared desirable; more greater speed, easier programming and greater reliability.
Electronic speed had enabled small jobs to be done very quickly and made possible computations which had up to then been unthinkable. Greater speed would enable a more rapid throughput of jobs - a practical aim which Williams and Kilburn always had in mind - and would open up new markets, particularly at that time among aircraft designers who were calling for bigger and bigger matrices to be manipulated and in important "real time" operations. Weather forecasting by computer might become possible if the differential equations could be solved within hours. There was also a growing interest in linear programming, particularly for the solution of transportation and "least cost mix" problems. A quick look at one of the transportation calculations - for Scottish Airways - had shown that at Mark 1 speeds of computers the problem would have to be so much reduced by eliminating the less important variables that, given a few more assumptions a desk calculator would suffice. There was some argument from potential customers that the attainment of higher speeds was an engineers' fancy, but the experience of users quickly showed up its advantages and that, given reasonable reliability, speed became perhaps the strongest selling point at any given price level of computers. It was of course the easiest attribute to discuss.
The second point was the introduction of a floating point accumulator. This would increase the speed of additions and subtractions in floating point twenty-fold compared with the programmed operation in the Mark 1. It was calculated that at comparatively small extra cost - less than 4 per cent - automatic floating point could be provided. It was bound to increase complexity and part of the purpose in introducing it into the design was as an experiment to see the effect on maintenance costs and reliability. Also a floating point machine would be slower than a fixed point one on fixed point work and to the skilled and experienced programmer the advantage of floating point was not great over a range of jobs. The aim, however, was to widen the use of computers and this meant encouraging workers who were experts in many fields of study to write their own programmes. It could not be assumed they would become expert programmers, so it was essential to make programming easier.
Thirdly, the reliability of the computer had to be higher than of the Mark 1*, if possible. If the percentage of useful time in a given period were fairly constant it would be possible to organise the work to allow for it, but the distribution of useful time is unfortunately very uneven and the effect on jobs can be very serious and the waste of time by users who have to travel long distances to the computer very frustrating, if they planned to rely on it and it failed on the day.
The idea of a floating point accumulator was exciting to Lord Halsbury and he put before the Brunt Committee a proposal to support this innovation.
There was then the possibility that N.R.D.C. or D.S.I.R. would provide funds. Ferranti had applied to the D.S.I.R. for £25,000 to finance a programme and the application was considered by the Brunt Committee. N.R.D.C. were willing to support the manufacture of the Mark 2, but after protracted negotiations their offer was rejected. Sir Vincent always regarded their terms as unfavourable and Pollard persuaded him that the computer would be cheap to make and should be very profitable. I advocated staying with N.R.D.C. because of the insurance they gave against loss and their backing had undoubtedly been useful in selling the Mark 1*, they represented a Government commitment and so gave confidence to possible purchasers who were working in government establishments or elsewhere on government supported activities. The Mark 2 was likely to be used in these kind of establishments.
From the sales point of view we were not enthusiastic about the proposal to make Mercury (using the name given to the University Mark II by Ferranti). The price of £80,000 or more plus heavy costs of maintenance had been a considerable deterrent to sales of Mark 1* and we were anxious to get a computer to compete with the Elliott 401 and also with the English Electric Deuce which was known to be coming fairly soon. DEUCE seemed to be a powerful threat. This seemed to mean aiming at a selling price between £25,000 and £40,000 and Halsbury was anxious that a computer derived from the 401 should, if possible, be priced below £30,000. This was more attractive than a big computer.
After a meeting in the office of N.R.D.C. I wrote that I could not see sales of more than four of the projected big computers - though it was of course early days. Some years later one of my programmers, John Davison, told me he had come across this memorandum; we had then sold about a dozen Mercury's. I said it only showed how difficult the market prophecies could be, but John then said that apart from the sales to nuclear establishments we had in fact sold four. Ultimately we sold 18 apart from the computer for Manchester University: 10 for atomic energy work and 8 for other purposes, including two big systems to Shell and to B.P. for linear programming work, 2 to the Universities of Oxford and London, one for weather forecasting, one each to I.C.I., R.A.E. and Metropolitan-Vickers. We should perhaps have foreseen the atomic energy market, but I do not think that at the early stage we had confidence that the big computers would really be reliable enough, and it did take a long time to attain a desirable level of good operating time.
It was the high speed of operation which gave Mercury its advantage, plus the fact that though there were larger as well as smaller computers in the U.S.A., none there was as good value for money.
The designer's problem was how to achieve a high speed without too much complexity and consequent risk of unreliability. A basic increase in speed over the Mark 1 could be obtained by operating the control and accumulator sections in a serial manner at 1 Mc/s. This involved making a store which would provide and accept serial information at a digit repetition rate of 1 Mc/s with immediate access. At that time magnetic core stores were only just coming in, whereas the University had had a lot of experience of cathode ray tube stores. Both had an upper limit of operating speed of about 100kc/s. It was therefore decided that ten stores of either type arranged in parallel might be used. The ten inputs or outputs could then be combined to give or accept serial information along single channels at a digit frequency of 1 Mc/s. The University machine was first made with c.r.t. stores and the Mercurys made in the Ferranti factory had core stores.
This immediate access store of the Ferranti Mercury was arranged as ten planes of 1024 magnetic cores 2 mm in diameter, each arranged in a 32 × 32 matrix. The speed of the magnetic cores was such that the ten bits could be stored or extracted in 10 microseconds. The digits as they were read were inserted into a serialisor or read unit emerging as a 10-bit serial word.
These 10-bit or "short" words were used in connection with a B store and a short accumulator, the B store consisting of eight registers each of which was a serial 10-bit electromagnetic delay line store. One of these registers was given special facilities to enable B-modified B-instructions to be carried out.
A complete instruction required two 10-bit words. The first specified the operation to be performed, or present function (P.F.), the second, call the present address (P.A.) , specified the location of a number to be operated upon or was itself an operand. Long numbers consisting of 4 short words (40 bits) were used in the arithmetic unit.
Since it was a floating point machine numbers were specified in "floating binary" form, ie x 2y. x had 30 digits and the exponent y, 10 digits. The x's were considered as lying in the range -1 < x < 1 - 229, a "1" in the most significant position representing -1 and the other positive fractions and these numbers were normally stored in a standardised form.
The times for various types of operations were:
Time for Time for fixed point floating point version version B-register and control 60 µs 60 µs Addition and subtraction 120 µs 180 µs Multiplication 210 µs 360 µs
These figures can be useful for making comparisons but of course the actual time taken depends not only on the time to perform the function but also on the time taken to transfer information between the main and subsidiary stores and to get it in the right place inside the computer. Evan with the longer floating point times this still applies.
The large-capacity backing stores were of the type of magnetic drum which had been used with Mark 1. Each drum had 64 tracks and after allowing for parity check and gap digits the useful content of each track was 2560 digits. The first machine had 4 drums giving a total of over 600,000 digits.
The construction of Mercury was in physical sub-units called chassis and each of these represented some major sub-division of the computer. Although the electronic design of the various logical devices was approximately standardised there was no attempt to make the physical partitioning of the assembly correspond with the conceptual partitioning into small groups of gates; the chassis were therefore all different. This gave some trouble in maintaining the computers in the field and the inevitable differences in the numbers of the various types of peripherals which were added increased the complexity.
Mercury was begun as a rather ambitious further experiment in computer design and in Ferranti we ought to have given more man-days to the re-design. But from being concerned as to how we should get orders we quickly passed to a situation in which we could not meet delivery dates. Pollard felt strongly the need for the Manchester team to beat the London Pegasus one.
This situation was exacerbated by the need to add better input-output facilities. Paper tape sufficed to prove new engineering developments and for a very large number of scientific jobs. Perhaps it was adequate for most of the University computing jobs. But when potential customers considered the computer -and at that time they had to be big customers to consider it - they often wanted to use cards because they had accumulated them in their systems. Here again the tie up with Powers was not helpful: the Sales Department was not supposed to be concerned with design information - it was the province of Welchman - and the Powers Organisation did not bring us into contact with the customers who would have helped to divide a suitable card input to Mercury. One result was that later elaborate and expensive punched card units had to be designed for the oil companies. We had a machine which was very useful for their large-scale linear programming work, but did not have, until very late, equipment to feed in punched cards. When we did valve machines were outmoded.
So far it has not been possible to find out what happened in the design team of Welchman and Whitwell. Even Peter Simpson, who worked for Whitwell and is now of I.C.L., does not know because, he says, secrets were as closely kept within Powers as they were preserved from discovery by outsiders.
In 1970 four of the nineteen Mercury computers were still working; one at the French Atomic Energy Establishment installed in November 1957; one at C.E.R.N. in July 1958; one at the Belgium Nuclear Energy Establishment in September 1959 and one at Buenos Aires University in October 1960. These could clearly have brought much more than their sales price if they had been rented.
The Atlas Computer was built by Manchester University and Ferranti after the efforts of N.R.D.C., the Ministry of Supply and the Atomic Energy Authority had failed to find a way of answering the threat of the big American computers. Had they dome so and provided an environment and the means by which so ambitious a project could have been completed quickly and launched in time to capture an important part of the world market for large computers Atlas might have had a much greater success. It turned out however to be a bigger job, particularly in the making of its software, than was anticipated and perhaps bigger than could possibly have been foreseen. It has to be remembered that I.B.M. was unable to envisage the requirements in their parallel and comparable exercise STRETCH and this was a failure. So was the Bull Company's Gamma 60. By comparison Atlas was a success.
One of Christopher Strachey's recommendations on his return from Canada at the end of 1962, had been that a design should be made for a powerful mathematical computer. N.R.D.C. had therefore offered to support Ferranti in making the Mercury computer and had supported E.M.I. in making the 2400. More and more powerful computers had been emerging in the U.S.A. and were promised from Europe and it was towards the end of the 1950'a that the powerful STRETCH was being made by I.B.M. and the Gamma 60 by Bull in France. Lord Halsbury therefore thought it was urgently necessary that the U.K. should have a powerful contender in the race. It can be argued that this should not have been done without a research into the market but had such a research been made the computer would have been stopped, almost certainly. Instead the Very High Speed Computer Project was launched. This initial research should have been Ministry financed and N.R.D.C could then have financed the commercially viable model based on the research.
The problems raised by Atlas and which led to it not achieving the success it might have had were: lack of funds; lack of confidence; failure to realise the value of an ATLAS-based network; the questionable value of making the ATLAS 2 design; and the doubtful wisdom of the decision to withdraw it from the market in favour of the 1906A. Also when so big a leap forward was being taken it would probably have been wise to make one - or perhaps two - before freezing the design for production. This would have been costly but no more so than the cost of making both Atlas 1 and Atlas 2. We had seen the advantages of designing Mark 1* after some experience with Mark 1.
In the British Computer Industry lack of funds to launch new machines properly had to be accepted in the early days. For ATLAS all sources were tried and were found unwilling. Lack of confidence that Ferranti could do the job was understandable and prevailed in official circles. To begin with, the fact that I.B.M. and Bull were making very big machines meant there would be competition for a probably limited market and when these two companies were seen to be in difficulty, it was only to be expected that the Ferranti project must run into serious, expensive and perhaps catastrophic trouble. The weakening competition weakened the market.
It was unfortunate that the idea of a network of ATLASES was not seized upon in this country. When Mr Sebastian de Ferranti referred to it in his address at the luncheon to celebrate the A.E.A. order, it only provoked laughter among the distinguished scientists and others present. By contrast when mentioning at a meeting at Edinburgh to Sir Fred White of the C.S.I.R.O., Australia, he immediately seized on it and at Canberra it was not difficult to convert the experts who had been studying the introduction of two machines, one for scientific and one for administrative work, to acceptance of the value of the two ATLASES, backing one another up at the centre of a network. But of course a network needed dispersed terminal facilities and as our designers had not been thinking in network terms, these were not available. C.D.C. had them and our efforts with C.S.I.R.O. turned to their advantage.
Whatever the relative merits of ATLAS 1 and ATLAS 2, the work created by the two designs was too heavy and the software which in any case was falling being was further delayed. The problems created by having to maintain ATLAS 1 as a production machine were increased by the removal of the Mercury computer from Manchester University while the Atlas software was still being developed.
The question of continuing ATLAS or not did not arise until Ferranti had been taken over by I.C.T. but it is questionable whether, rather than replace it with 1906A, it would not have paid to keep it enhanced by using more modern components which could, I understand, have increased its speed 6 or 8 fold.
The very high speed computer and the A.E.A. requirement were not necessarily the same. The former was envisaged as leap-frogging over existing designs rather than fulfilling a specified need, such as the Authority would have. Events brought the two requirements together.
In 1956 a project for a large computer was proposed by Sir John Cockroft and in January 1957 Howlett wrote to him indicating the requirements in such a computer. In April the first Harwell Conference was held and the outline specification was discussed. It seems to have been agreed that the need would be for a speed of about 109 operations in 15 minutes - the word operation not being precisely defined. Since 15 minutes is 900 seconds this simple formula leads to a speed of the order of 1 µs per "operation".
In May the very high speed computer was discussed in N.R.D.C. against the background that the Atomic Energy Authority might have to purchase an I.B.M. STRETCH pending the making of a suitable computer in this country though there was a desire to support the project in the U.K. because of the help it would give to the computer industry. In June, noting that I.B.M. was reported to be spending $28M a year on R. & D. and was said to have 300 graduates on STRETCH alone, the N.R.D.C. Board was asked by the Computer Sub-Committee to approve expenditure of £1M. They noted that a number of Government departments and other organisations were interested in the High-Speed Computer Project. T he U.S. : U.K. ratio of about 10:1 is seen, as so often in comparison of U.S. : U.K. activity and resources, to come in here.
The problem as to who should assume responsibility for the project was still not settled at the end of 1957, N.R.D.C. wanted to encourage it; there was a suggestion that the Ministry of Supply should provide the funds and Sir Owen Wansborough-Jones said that he might take on the job, with reluctance, at R.R.E. Malvern.
In February 1958 a technical meeting at Harwell heard and were much impressed by F C Williams' proposals which were for a computer 50 times faster than Mercury and perhaps one-third to one-quarter the speed of STRETCH. (By 1959 the ratio of power was more like 1:2, but this may have been because of a weakening in STRETCH which in due course was to be found much less powerful than planned). Sir John Cockroft asked whether the A.E.A. should not support its development but Sir Edward Plowden, Chairman of the Authority, said this was not possible, though they would need a powerful computer by 1962. By June it was still not settled who would do the job. Cockroft wanted Manchester University and Ferranti to go ahead on lines as proposed by Williams and Kilburn. This produced nervousness on the part of Wansborough-Jones and Halsbury. The former was dubious about Ferranti being able to afford to undertake so big a job and Lord Halsbury was worried about the domestic difficulties which might arise between the engineers and the programmers. He had seen this before in the University-Ferranti collaboration and his concern was very understandable. There were to be difficulties, but on Atlas they were probably, more than on most computer projects, due to the inherent difficulty of the job rather than to personalities. In so big a job there are an enormous number of decisions to be made and comfortable compromises are not easily achieved.
Because of the doubts English Electric were approached but do not seem to have been enthusiastic. At the beginning of 1959 Ferranti put forward proposals based on Williams' ideas but N.R.D.C. could not accept them. E.M.I, also put in a bid in which they could see a machine emerging which would sell for £300,000 to £400,000.
In March 19S9 the N.R.D.C. computer sub-committee recommended acceptance of the E.M.I, design, which they thought had the advantage by a narrow margin. This recommendation was not accepted by the main Board of N.R.D.C., who preferred a proposal by Dennis Hennessey that both E.M.I, and Ferranti should be supported by grants of £250,000 in the case of E.M.I, and £300,000 to Ferranti. E.M.I. accepted but Ferranti said they wanted to talk to I.C.T. and E.M.I. It was presumably the terms proposed by N.R.D.C. which led Sebastian de Ferranti to say "we had thought of calling the computer BISON - built in spite of N.R.D.C."
These discussions therefore merged with the talks which had been going on at Sales level about the possibility of co-operation between the engineering companies. The various negotiations came to the point where a company jointly owned by Ferranti, E.M.I, and I.C.T. was proposed. Sir Joseph Lockwood says that these failed because Ferranti felt that as a family firm they could not join with two public companies. This would have involved more separation than Ferranti would have liked, because the organisation would have had to sell some or all of their computer interest to a separate company, but this was done later when Ferranti sold out to I.C.T. and to have sold earlier could have preserved the Ferranti leadership. It was perhaps unfortunate that these top-level talks brought about by ATLAS problems were not discussed along with the reasons why the same three companies were studying at sales manager level to need to amalgamate or at least co-ordinate their effort. The greater engineering strength might have provided the answer they were looking for as well as speeding up ATLAS. E.M.I, joined with I.C.T. and from this point on to the present day the British Computer Industry was to be dominated by I.C.T. ATLAS, which with the failure of STRETCH could have become the world's most powerful computer, had a small patronage and the market for which it was made was largely taken by the Computer Development Corporation.
C.D.C. afterwards said that the talk given by Kilburn in 1959 in Paris on the Manchester University project had saved them a great deal of time and money because it told them of avenues which had been fruitlessly explored and stones which were not worth turning over. This may have produced one of the turning points in this project, because in 1953 when the Australian Government officials came to decide between C.D.C. and Ferranti, the former were able to show a fully working system with compatible medium-power computers while the Manchester computer was still not finished. At that time Australia was the best market for ATLAS.
Atlas was designed to meet the needs of the large scale computing and data processing centre. There was now no talk of a scientific or data processing machine, but of a computer to handle the widest range of work from high-speed computation to routine data processing at low cost per operation.
Manchester University and Ferranti designers set out to meet the criticisms made of Mercury and that its input/output facilities were inadequate and to ensure that it could have the maximum throughput. The Atlas system was designed with a hierarchy of stores with different access times and facilities to allow simultaneous operation of several programmes, each with its own peripheral devices. The order code, of single address type, is relatively simple, providing for all the elementary operations, in both fixed and floating point form. But in addition, a number of additional and more sophisticated commands - called extracodes - were made available by using a fast read-only store of novel design. It was the description of this store which led some competitors' salemen to express their envy at the Paris conference in 1959, the prelude to the formation of I.F.I.P. They had asked for the facility and been told by their engineers it was not practicable. This store was constructed from a woven wire mesh into which ferrite rods representing digits were inserted. The access time was 0.3 microseconds and in the original machine at the University it had 8192 words although this could be extended up to a theoretical maximum capacity of 262144 words. This fixed store uses a subsidiary core store as working space for fixed store routines.
Next in speed is the B-store which contains 128 half word (24 bit) index registers to which access may be made in 0.35 microseconds. Here, as the arithmetic unit is distinct from the floating point accumulator, indexing commands can be executed while a floating point operation is in progress.
The main store has a cycle time of 2 microseconds, but this time is effectively reduced by the provision of a number of distinct access systems to sections of the store. The first machine had 16384 words in the core store and 4 access systems. These independent systems were so arranged that consecutive commands were read from the even and odd registers in the store through separate systems, so that while one command is using the accumulator, another is being completed by sending information to the store and yet a third is being initiated by sending the command from the store to the control.
There are also a number of registers, private and accessible only to the fixed store routines, which are used for the information and control signal passing to and from the magnetic drums, magnetic tapes and other peripheral equipments. These are collectively known as the V-store.
The apparently complex design becomes much more simple when looked at from the user's point of view. The "private" stores should be considered as carrying out some of the functions of the control unit and the user does not need to be concerned about how the engineers have organised this. Also, the core store and the drum store are together organised in such a way that they appear as a single level main store to the programmer. He can, therefore, look upon the heart of the Atlas computer as having a main store and a control which is connected to an accumulator, and a B-store with its own arithmetic unit. The operating controls and the input and output units are connected to the control and the magnetic tapes to both the control and the main store.
The unification of the main store (core and drums) is accomplished by means of sets of registers in the V-store called page address registers. Information in the store is regarded as contained in a series of numbered blocks, each of 512 words. These blocks are transferred between the core and drum stores from time to time and a list of blocks currently in the fast store is contained in the page address registers.
When a command is to be obeyed the page address registers are scanned very rapidly by special hardware and if one of them contains the required word, it gives a recognition signal and the corresponding page in the store is used. Therefore, when the particular command and the required data are in the store, access is made at the full speed of the core store. But if a word is called for which is not in the core store control is transferred to a special drum transfer routine.
This routine selects a block in the core store which is to be replaced by the required block on the drum and arranges to write the replaced block on the drum and enters in a directory in the subsidiary store where it has been put. This saves time because a block can be put into the next available space to reduce waiting time. To enable the drum transfer routine to do this the current angular positions of each drum are kept available in registers in the V-store.
The user therefore appears to have a very large store much more cheaply than if magnetic cores were used throughout. It has to be remembered that discussions about the design of Atlas had to be made in many cases, before 1960, and very large core stores though becoming cheaper, we still a luxury. Time sharing of a number of programmes, without fear that an error in one will interfere with the others, is controlled by supervisory routines in the fixed store.
ATLAS was described by its designers as having "a large quantity and variety of peripheral equipment for input and output"; but this was written in 1961. In 1972 an article in the Harwell Report on Research Application said "(ATLAS) is not not readily suitable for inexpensive attachment of a large number of input/output devices". Over the twelve years the need for a multi-access system had grown, calling for teletype terminals and other slow speed devices for which the ATLAS interface, designed for rapid exchange of large quantities of information and to maximise the throughput, was not designed. So the designers of the system were virtually certain to make some decisions which would later give difficulty.
Professor David Howarth, looking back at the programming problems which had to be solved says they fell into three groups: First, techniques had to be developed for dealing with parallel activities and giving the necessary interrupt facilities. As the problems arose ad hoc solutions had to be found. Second, it was inevitable to keep an understanding of what was being done, that in the early stages low level techniques had to be used. Then once the probing stages were over, part of the team had to be put on to improving the tools which had been made. This took a long time and it was the mid-sixties before really satisfactory production tools could be perfected. Third, there was a further strain put on the programming staff when the University decided in January 1963 to get rid of their Mercury computer. This meant that effort had to be diverted to keep ATLAS fit for production work at the expense of the construction of the supervisor software by which the computer did its housekeeping.
The whole job took a long time and the ATLAS work was included in the general castigation of manufacturers' software which has been made from time to time. This was clearly expressed by Professor Michaelson in his paper given to the I.F.I.P. Conference in Edinburgh in 1968; In Universities, he said, "things are done by small groups .... manufacturers are bemused by their own sales talk .... when things fall behind a few dozen bodies are added .... where a user group has perhaps half a dozen people, the manufacturer has a couple of hundred". We can claim that in Ferranti we tried to avoid the enormously expensive and almost unattainable team Michaelson criticised. Later that year he made a handsome withdrawal of his criticism so far as ATLAS was concerned.
The organisation of this programming work was begun by Dr (later Professor) Stanley Gill who was later succeeded by Hugh Devonald, but ultimately as the work proceeded and Devonald had to move to other work, the load fell on David Howarth and his team.
Dave Howarth was one of the most extraordinary workers in a field which has produced a number of men and women with remarkable mental and physical capacity. Physically of medium height and spare build, he could continue at a level of activity which would have exhausted most men. He had gone with a scholarship to Imperial College at an early age, renouncing a Cambridge scholarship because he would have had to wait a year before he would have been allowed to take it up. By the time he was 21 he had a first class degree and a PhD. He had been working at R.R.E. for some years and when he applied to us he was clearly God-sent material for Atlas. Round the University programmers and Howarth we were able to build the sort of small, highly skilled team we liked to have.
There is one problem in designing computers and putting them to work which is continually troubling the engineers and the programmers. The programmers cannot test their programmes without a working computer and the engineers cannot progress until their current work has been tested and approved. It is very frustrating and calls for more restraint than is usually available in human beings. It was better on the later machines than on the very early ones like the Manchester Mark 1 because then it was very often relatively elementary things that went wrong. By the time of Atlas the troubles had become very sophisticated ones and these were easier to live with in spite of the frustration of the would-be users.
Could the whole exercise have been speeded up? It took nearly four years from the time when ATLAS could be said to be working on production jobs to the time when the software tools were "perfected" and this was ten years after thinking about the computer began. History may find an answer to the question, as history usually does in the long run, by losing so much of the detailed information that the answer is to a question different from the one originally posed.
One of the remarkable features of the ATLAS project was the small number of staff employed compared with the large numbers used by I.B.M, on STRETCH and by the Bull Company on the Gamma 60. We have seen that, probably, the former had some 300 graduates and the latter about 200 programmers. ATLAS never had more than ten programmers on the supervisor and about 15 working on compilers. The total number of engineers on one computer must have been of the same order. To get exact and comparable figures is probably not possible; Ferranti resources could not without external assistance have maintained the sort of production and sales forces which the punched card companies conventionally considered appropriate. As before our method had to be selective, and small-scale; we tried to find programmers who could really do the job and to avoid their time being wasted by large numbers of unsuitable assistants.
We can now look back to see what can be learned from the exercise. There do not seem to be criticisms, nor more surprisingly, self criticisms of the decisions taken about software, nor were serious shortcomings found in the architecture of ATLAS. It was however, not appreciated until everyone was too much involved to stop, how the big job was not how much of a research project had been taken on. This meant that timing could not be properly estimated.
The answer was surely that Ferranti were really right to consider that two of those advanced machines - one for the University and one for the Atomic Energy Authority - should have been made before a production model was decided upon and financing became dependent on sales. This would probably not have been too late for the market (Dave Howarth mentions that by 1967-68 ATLAS could have become six or eight times as fast by using modern components).
Was then ATLAS too big a leap forward. Was Strachey right in thinking that the E.M.I, proposals were a better answer to the immediate requirements of the early sixties? and the N.R.D.C. Board right in wanting to support both E.M.I, and Ferranti. We can say that the scale of the support N.R.D.C. were proposing was too small - by nine-tenths or more - but it could perhaps have made sense for the two companies to work on both machines. The E.M.I, development of their 2400 would probably have been easier to make and could probably have held the market until ATLAS was perfected. But we did not then know that I.B.M. and Bull computers would be such failures and in any case the competition of C.D.C. was coming up. We did however know that the effort we could put on the Ferranti computers was not enough.
The essential problem is the uncontrollable rate of advance of computers. Other products can be better disciplined.
A few years ago I met Dr Pope, then Director in charge of Diesel development at Brush. I had known him when we were both sitting on an ad hoc committee at Nottingham University where he was then Professor of Mechanical Engineering. He told me that he did not allow an engine development to go ahead if it aimed to get an advance of more than 10 per cent on present achievement. When we see the speed of development in computers we get an idea of the difference between electronic and mechanical engineering.
After our experience with Mercury, the two most obvious markets were Universities and nuclear energy establishments. Manchester represented and in the first instance might serve the former. Harwell was the obvious nuclear customer. London University was the most promising of its group. We hoped we should not again have the experience of getting our first order from abroad.
The first stage was however the clarification of the position of the Very High Speed Computer. The promise of STRETCH by I.B.M. seems to have settled this issue and work was concentrated on developing a suitable machine for the A.E.A.
In 1961 the U.K.A.E.A. ordered ATLAS and this order was celebrated at the luncheon held in the Savoy Hotel mentioned earlier to which a number of possible potential customers had been invited.
In Europe there were several nuclear establishments which had bought Mercury computers and of these CERN certainly seemed a likely customer.
There were obvious potentialities in the U.S.A. though R H Davies, perhaps because of his experience with Mercury and knowledge of the difficulty of selling any European computer there, was not very encouraging.
In 1961, Hall, Gill, Hunt and Swann went to the U.S.A. and talked to a number of potential customers. There was obviously a great deal of interest though also some doubt about whether the project could be realised. It was decided to send John Fotheringham to spend some months there and he pursued the original and other possibilities.
The other interested market was Australia, where in 1961 it had been decided to open an office with Barry de Ferranti in charge. We had looked at the market there in 1959/60 and the country was clearly going to be an important and moderately big outlet for computers. The size of the country relative to the size of the population and the need to establish control of Government computing activity at Canberra were important factors. Growth was bound to be rapid and a sophisticated attitude towards computers prevailed. I had an opportunity to tell a party of Heads of Commonwealth Scientific and Industrial Research Departments in Edinburgh that our idea would be to have an ATLAS at the centre of a network connecting to peripheral points. Sir Fred White of Australia was immediately very excited, when I went to his country in 1962 it seemed almost certain they would want ATLAS for the CSIRO, located at Canberra. Barry had worked very hard and persuasively and the general tone of the conversations when he and I visited the interested people was expressed by Sir Leslie Martin at Melbourne: "when we have the ATLAS at Canberra". There was a discordant note struck by Dr John Allen Ovenstone in charge of the Defence Department computer. He had seen details of the C.D.C.6600 which he described as a "beaut" and he introduced me to a staff major from the Department that same evening as "in charge of marketing for a company that cannot deliver the goods in time". These two remarks made during the hour before dinner in his house were chilling. There was however some compensation from the administrative side of the Government.
For several years they had had a Civil Servant, Barry Pridmore, studying the needs of the Department of Commerce and, at the same time as the CSIRO called for tenders for an ATLAS-like machine, invitations were issued for tenders for a commercial machine. Harry Johnson was quite unable to cope with the demands for studies of applications of ORION and we were running into difficulties with the computer so I asked to be excused from quoting. This was a great disappointment to Barry Pridmore but overnight I had the idea of asking them to use ATLAS for this work. The acting Commonwealth Statistician in the absence of Sir Stanley Carver was Keith Archer. I had met Carver in 1936 when he was visiting London with Mr B S Stevens, the Prime Minister of New South Wales, and I had done a report for them on the marketing of Australian farm products in Europe. We had kept in touch but unfortunately I found he was now away so I spoke to Archer. He was immediately attracted by the idea of two similar machines, the programming and maintenance would be eased and the computers would back up one another in the event of breakdown or overloading. We looked to the prospect of selling 2 ATLAS computers to Canberra which would have enabled a proper organisation to be established.
In the summer of 1963 Dr Geoff Hill of CSIRO was sent to study the 6600 and the ATLAS. He was, I am sure, anxious they should buy a British but C.D.C. had, he claimed, got their computer to a more fully commissioned stage and they had the advantage of the 3600 computers to serve as satellites in the various State Capitals of Australia. Barry de Ferranti has also since said their price was cheaper. I have subsequently heard, though am not sure about the story, that C.D.C. had difficulties with installing and it is not certain the Australians were satisfied with their decision.
Fotheringham's efforts came nearest to success at Westinghouse. Dr Edwin Harder, who was in charge of the plans for a large computing activity in that company was very interested in ATLAS. Several years later he told me that their plans at that time (1962) would have given Westinghouse power to do, in theory, all the computing then necessary in the United States. Even allowing for the very big difference between theoretical and realisable achievement this statement indicates the speed at which computation was thought to be expanding. Edwin said they came near to ordering ATLAS but finally decided to increase their battery of I.B.M. computers. He seemed to imply that buying a foreign computer of which there could only be very few in his country was too big a deterrent. Also the ATLAS could not be seen working when his decision was made.
In 1962 Grundy pressed that we should make a further effort in the U.S.A. and we engaged David G White, then with Plessey, to lead a small team in a campaign. He reported in February 1963 on the following:
These seemed considerable possibilities and with sufficient sales efforts could perhaps have yielded orders in spite of U.S. reluctance to buy foreign. But ATLAS was not fully working? C.D.C. were coming along fast and I.B.M. were not likely to let the collapse of STRETCH defeat them. The ATLAS lead could not last. White had to be convinced before any would buy; none would want to be considered in isolation. They would need a lot of personal contact because they were used to this treatment and they would need to be convinced that Ferranti were of sufficient stature to do the job.
This was at a time when Ferranti were having to recognise that they were clearly not of sufficient stature without joining with others. ORION was being difficult and a lot of management effort had to be diverted to it.
There was another difficulty which arose out of the above. Since the Americans were not likely to move unless assured that several ATLASES would be in their country there was the danger that important penalty clauses might be demanded in contracts - and accepted. We were claiming that one ATLAS equalled three I.B.M. 7094's worth £1 million each. Half a dozen could mean putting £20 millions at risk if we were late in delivering. We had gone so far from the days of the computer which was going to cost £300,000! But the problems of punctual delivery were not yet solved.
In the U.K., C.E.I.R. set up a London Company and the people concerned were well known to me. Herbert Robinson, who had taken over the small moribund organisation run part-time by academics in Washington and quickly made it into a powerful business, was a Yorkshireman I had known from the early 1930's. A graduate of London University with a doctorate from Oxford, he had worked with Professor Lindemann during the war when we again came together and then gone to America where he formed C.E.I.R. Ltd.
In London he invited Or C O George, an old friend of mine and former fellow Chief Statistician of the Board of Trade, to be the Chairman. They visited me in Portland Place to get advice about a possible Managing Director. Amongst others we discussed A S (Sandy) Douglas, a former E.D.S.A.C. user at Cambridge and at the time in charge of the computing at Leeds University on Pegasus, and Tom Cauter, who I had first known in 1947 when he was Managing Director of the British Market Research Bureau, a subsidiary of J Walter Thompson. Tom had gone to America but Oswald George got him back to become the Managing Director of C.E.I.R.(U.K.). He had been one of the first people to be interested in the possibilities of the Mark 1* and was now very much attracted by ATLAS. He did in fact have the first provisional quotation which was for about £1.1 millions for a minimum sized computer.
Cauter persuaded Professor M G Kendall from the London School of Economics to join his Company and this move linked with British Petroleum provided part of the money for the ATLAS for London University.
We had hopes that they would buy an ATLAS. We had engaged Dr Aylett to specialise in selling our computers to the C.E.G.B. and, in due course, with some success. But, although Hawkins there was interested, he remained reserved and continued to say he could not justify ATLAS.
We hoped to have I.C.I, as a customer but they had to have satellites. G E Thomas said that if we could not provide these "I.C.I, would never buy an ATLAS nor would they use it much". He wanted the first satellite to be available at Newman Street. Although we got on well on the control side, I.C.I, continued to be rather unforgiving about our interest in Powers cards which they considered interfered with work on the Hollerith version.
The computation work did not build up in Geneva as rapidly as had been expected.
This sample of experiences up to 1963 show there was a fair market and the technical success of ATLAS suggests that a fair sizable and perhaps big market could have been made for it. But When I.C.L. bought the Ferranti Computer Department in 1963 they had to decide which of the numerous computer they now possessed should be continued and which should tail off. They decided to concentrate on the 1900 series derived from the Ferranti F.P.6000 which was in turn part derived from ORION. Tha success of C.D.C. in the big computer field suggests that it might have paid to keep ATLAS in the market, particularly in view of the increase in speed it could have achieved with modern components and its sophisticated software.
It seemed right that Cambridge University Mathematical Laboratory should have a powerful computer, so one day I rang Maurice Wilkes there to ask if I could see him about ATLAS. He invited me to dine with him in Hall but when I came to enter the date in my diary I found that it was Whit-Monday. Before we could fix another date Peter Hall had spoken to him about the possibility of letting his laboratory have ATLAS hardware on special terms in return for the design of a modified - and hopefully cheaper design. This was made but the changed hardware meant more software and late deliveries. For several years I was dogged by this machine because the Under-secretary at the Ministry of Technology who was responsible for the first installation in a Government organisation would not accept that Ferranti Ltd connections with job had ceased several years before.
Soon after the launching of the Mark 1* computer, Gordon Scarrott began working at Moston on a "Neuron" device. This was a transistor equivalent of a logical device using magnetic cores which depended upon an output resulting from the algebraic sum of the effects of positive and negative inputs. This was seen as another way of making standard logical devices which could be put together by logical designers in much the same way that the Pegasus packages were organised, but in this case by using transistors were regarded as exotic and expensive and the neuron designers decided they should maximise the logical power attributable to each transistor.
This was a mistake, but understandable. It is hardly part of a research physicists responsibility to predict the probable future trends of prices. It was a fault in management which kept projects so secret that quite elementary facts - in this case quite elementary matters of economics - were not discussed. New devices with high R & D content are expensive, but small devices like transistors will either always be difficult to make, in which case they can serve only in a limited market, if at all, and we should have looked to other techniques for moderate priced computers, or ways of making them on a larger scale will be found and they become cheap. It would be difficult to guess when the drop in cost would come so the proper course would have been to do the preliminary R & D and so be prepared to take advantage of lower costs if and when they came. Manchester University showed a clear appreciation of the economic logic when they began to use transistors, they chose the best they could find for the job, confident that when needed in quantity, the price would have come down. But their plans were for advanced computers which would take a fairly long time to complete; the problem of suitable components for more modern machines is more difficult.
The aim to use each transistor to the full was achieved only at the expense of tight timings and component tolerances and it was perhaps these characteristics which made some of the Pegasus engineers suspicious of the soundness of the neuron logic and communicate these suspicions to the Sales Department. Pegasus packages had been designed to give good safety margins. However, a small machine was built to test the neurons and the development team claimed that its performance demonstrated a much higher reliability could be obtained than by using valves.
A different circuitry was meanwhile being evolved at Wythenshawe by Maurice Gribble, formerly a member of the Computer Department engineering staff, and it was not long before sides were taken in favour of either "Neurons" or "Gribbons".
The fading hopes of getting anything from the Powers talks presented the Sales Department with a dilemna. Pegasus and Mercury computers were beginning to be delivered to customers from 1957 on, the Computer Exhibition was due in November 1958 and Ferranti Limited, the country's unquestioned leaders in computers, looked like having to talk only about computers which had become familiar over the last three or four years. We badly needed transistor computers and we should look out of line if we could only offer core stores in the big machines.
At this time it was realised that computers had great potential for commercial work. Others were designing such machines and Ferranti must not rely on the Powers designers.
Not everyone liked this. There were good arguments for limiting our business to scientific computers and we flirted with this policy. It was, however, defeated by the American tax system. I.B.M. offered big educational discounts, the figure quoted to us being sometimes as much as 60 per cent off the selling price. This was a particular blow to Ferranti; we had done well in the University market and there were advantages in concentrating on this kind of work, but the price of the main competitor to Pegasus - the I.B.M. 650 - made competition impossible when the customer could claim an educational discount from I.B.M. Furthermore, by our arrangement with Powers we had committed ourselves to making computers for industry and commerce, and although they were to take most responsibility for selling we should have to play a part.
The studies we had made in conjunction with Powers as well as on our own had given us a good deal of commercial computing knowledge. We had made the card converter-cum-line printer machine to go with a Pegasus system at I.C.I. Blackley.
Quite early on we had noted two characteristics of computers which were unexpected. When the operations to be performed were analysed it was found that the difference between scientific and commercial computing was much less than had been supposed. We had begun with the idea that scientific calculations involved complicated operations on a few numbers while commercial work consisted of simple operations on a large amount of data. In practice a lot of the work concerned with technical and scientific work also involved a lot of numbers; experimental data and observational results in physics, engineering and meteorology and so on. Also in all forms of computer calculations most of the time is spent in moving numbers about the machine and only a small proportion in actually performing the operations of arithmetic. About 80 per cent of the operations were organisational in scientific work and 90 per cent in commercial.
The difference between the two kinds of work were therefore much less than the similarities. But the sorting and collating of large numbers of punched cards and lists of names, addresses, product description and so on did present a problem which called for special techniques of economical storage, and magnetic tape was the medium used by Eckert and Mauchley in their original machine specially designed to do large scale commercial work. Drums gave quicker access but were too expensive to be provided in large numbers and the cheaper discs were not yet available.
Orion seemed to be the best answer we could be given to meet our own needs for a commercial computer system and it was very much welcomed by the Sales staff dealing with these problems. For the scientific market, in particular to meet the need for a replacement for Mercury, it was however too slow. At best it seemed to offer an increase of three or four times where we wanted ten times.
When, therefore, Pollard held a meeting in his office in September 1958 the arguments were concerned not only with the suitability of the neurons, which it was difficult to discuss because of the constantly difficult relations between the Manchester and the Bracknell teams, but also this question of speed. There was another problem raised; the possibility of a transistorised Pegasus. This was opposed by the argument that it would be very difficult to make, a virtual repetition of the objections produced when Pegasus was first designed.
The speed question was obviously not conclusive since for commercial work Orion was probably adequate and since the computer was to be extensible from a relatively small system (somewhere about £120,000) to a large one costing several hundreds of thousands of pounds, it could be looked upon as a suitable replacement for the Pegasus data processing system, though leaving us very vulnerable at the lower end of the market to Elliotts and, as we were to discover, to the I.B.M. 1401.
The meeting ended with Pollard saying that the machine was to be made regardless of the views of the Sales Department.
At the time we were too busy with preparations for the Computer Exhibition and selling Mercury and Pegasus to argue further, we had to rely on our sales programmers to make any system decided upon as attractive as possible. The Computer Exhibition came towards the end of November and at the end of the month Pollard resigned and during the following week Peter Hall was appointed in his place.
In the New Year the question of "neurons" against "Gribbons" was still not finally settled and I met Peter Hall in his office with Gordon Scarrott. Scarrott again hotly denied that there was any doubt that the neuron circuits would work. I suggested that as we had John Coales on the books as a Consultant it could be useful to have his opinion, but Peter Hall said this was unnecessary. He had himself examined the circuits and was fully satisfied that they would work. So arrangements were made for the design work to be put in hand: Dr Thomson led the engineering team and George Felton and Peter Hunt the programmers.
The design principles of the neuron was described in a short paper by Scarrott, Johnson, Haley and Naylor given at a meeting of the Measurement and Control Section of the I.E.E. on the 16th and 17th February 1959. It makes use of wound ferrite cores as linear transformers instead of as square loop storage devices. These transformers are employed to make a "ballot-box" logical gating. It was shown that a single gate circuit can suffice to perform the fundamental logical operations.
The pulses transmitted are current pulses synchronised to a standard clock frequency. These current pulses are summed on a transformer at the output to each element by using a number of separate primary windings, each carrying an incoming current in either a positive or negative direction. The circuit was arranged to generate an output pulse if, and only if, the total effective current at the primaries was positive. If the total current was zero or negative no output pulse was generated. An "or" gate was made by connecting all inputs positively to the input transformer so a current at any one insures an output pulse. For an "and" gate the inputs are also connected positively but, also a constant flow of negative pulses ensures that only when two input pulses coincide is an output pulse generated. An inverter is made by causing input pulses to flow negatively in one winding while a constant stream of unit strength is connected positively to another, so that an output pulse is only produced when the input has no pulse.
In the basic neuron circuit the standard current pulses last for one half of the standard digit time. The digit period is therefore divided into two equal portions each of one microsec in length, known as the charging period and the output period respectively. These are defined throughout the equipment by means of timing waveforms.
In the discussion on this subject at the I.E.E, it was pointed out that when driving along long wires it might be necessary to re-time pulses before further operations could be performed and the generation and distribution of high precision clock wave forms was one of the disadvantages. There was a story that when two of the engineers were working on the prototype Orion they found the currents at the end of the long lines very different from expectation and when they were satisfied that this was not due to careless misreading of the instruments, they did some study of the literature and found they were experiencing the Ferranti Effect, named after Dr Ferranti who discovered it. However, the neuron packages were incorporated in a test machine called NEWT and we were assured they behaved with high reliability.
It therefore seemed possible to proceed with the design of a computer which would, as far as possible, by time sharing different jobs, overcome the well-known difficulties and waste which arose from having to marry electronics to the much slower electro-mechanical equipments.
This was a problem from the beginning of computers. The paper tape moving at seven characters a second was a poor match to a computer capable of 1000 additions a second and efforts which raised the mechanical tape speeds to 100, then to 1,000 c.p.s. still fell further behind the more rapidly advancing electronic speeds.
The problem was therefore to find a way of using the time which would otherwise be wasted, e.g.that which elapses between the reading of one character and the next on paper tape or between the reading of two successive punched cards.
Evidently if the time could be so used we should make the computer operate on several times as many figures in a given time as before. It would then be best, indeed it would be necessary, that these figures should come from different jobs or different parts of the same job. So we could do several jobs at the same time and users would no longer need to await the completion of a big job before they could get on the computer to do a small one, or perhaps to test a programme. A new concept in operation of a computer to do a variety of jobs was opened up. The idea of a computer able to work steadily on a big job and within the same time fit in a number of urgent small ones was obviously most exciting.
Orion was therefore designed as a parallel computer with what was at the time a fast magnetic tape system operating at 90,000 characters a second, which enabled large files of business information to be processed at speed by a time-sharing system which controlled a number of card and tape input and output devices and enabled card to tape, tape to card, and printing from tape to be carried out without using expensive off line equipment. The arrangement of the store also permitted information to be transferred to it direct from the peripheral equipments, eliminating buffer storage. Lock outs protected each programme from interference by any other. A lot of attention was paid to this internal security.
The word length chosen was 48 bits representing fourteen decimal digit numbers or 8 6-bit characters. Since the computer was to be suitable for commercial type work special facilities enabled rapid conversion to be made from sterling, decimal or other radix numbers to the binary equivalents.
The various parts of the system were allotted priorities so that the slowest peripheral had preferential treatment to ensure that it was kept fully occupied. When this held up the computer the programme switched to the next equipment and so on. The time-sharer programme is part of the supervisory routine permanently stored in an isolated part of the store.
It was this facility which enabled Orion to do more work than the statistics of its basic speeds would suggest. These statistics are not simple because the computer had several forms of instructions. 3-address simple instructions took 64 µs, multiplication took from 156 to 172 microseconds.
It was intended that Orion should be an extensible computer so that a purchaser could begin with a small system and build it up as work demanded. Unfortunately when the design ran into trouble and there was difficulty in making it work, this not only made deliveries very late but led the engineers to design each computer to the specification given in the initial order from each customer, and the facility for expansion was lost or made very difficult. This was not learned by the Sales Department nor, it appeared, by the Departmental Manager, until too late.
By 1960 it was clear that Orion was such that it would be too expensive to meet the requirement for a successor to the Pegasus 2 Data Processing System and this as well as the growing engineering difficulties led to a suggestion that a "Pegasus successor" should be considered.
The specification was put in the hands of a study group chaired by Harry Johnson, who was in charge of the commercial sales organisation, and consisting of engineers and sales staff. This computer was specified and work was about to start in the autumn of 1961. However I returned from holiday in September to be told by Peter Hall that he had been in conversation with Ted Braunholtz, one of my programming staff who was working on Orion, and had been persuaded that it would be possible quickly to make an Orion 2 using the Gribble techniques Arthur Jackson, who had been called in to take charge of the Orion production, was doubtful whether with the neuron techniques, the machine would ever work, so Peter had already put the work on Orion 2 in hand and cancelled the Pegasus successor.
The Pegasus successor (nicknamed the HARRIAC) specification was later picked up by Ferranti Packard who made a computer very like it called the FP.6000. This became the beginning of the I.C.L. 1900 range of computers.
The organisation of the central computer to enable it to control a large number of peripherals and to share its processing time between several programmes required a sophisticated controlling programme. Orion probably had the first really comprehensive time sharing facilities so Felton's Orion Monitor Programme (OMP) was also a computer milestone. Meanwhile Peter Hunt and his team developed a new commercial high level language called NEBULA. The attractions of these two large software systems was sufficient to hold most of the customers we took for Orion, in spite of long delays in delivery, Only G.E.C. cancelled their order and as this came shortly after control had passed to Weinstock there may well have been some feeling of insecurity among the people concerned in the Company. Action would look better than acceptance of a situation which was certainly unsatisfactory.
It was necessary for G.E.C. to find another computer and their Computer Manager, Mr Lumb said when he told us of the cancellation, that he was unable to find a computer for their nuclear calculations better than the Mercury they were already using.
The instruction code of Orion - some orders being 2-address and some 3-address - gave the timings a wide range.
Fixed Point (µs) Floating Point (µs) Addition and Subtraction 36 to 68 90 Multiplication 60 to 192 180 Division 550 550
This language was developed by a team led by Peter Hunt because the existing languages were not considered adequate for use in the applications foreseen for ORION.
It was to be as simple as possible for the ordinary commercial use and therefore, although necessarily employing a rigid syntax, sentences should be formed as similar as possible to those ordinarily in use, the minimum of restriction should be put upon the user. Though data might be fed into the system from punched tape, punched cards, magnetic tape or other media, once in the NEBULA system all traces of its origin was lost. It could therefore use any card etc. code.
For effective file handling there were facilities for handling data files. In practice data items are of variable length so each record is fed into an input area which contains just one record and has to expand and contract to accommodate this. To identify data unambiguously, every record name and every file name must be different from every other data name of any sort on any of the files in the same programme. Each record contains a collection of numbers or of characters which is called, following punched card practice, a field. These numbers and fields relate to a "detail" e.g. item name, day etc. which must be clearly distinguished, but for convenience they often need to be described as a group as when "Data" means "day, month and year" and NEBULA enables the programmers and the machine to save time by giving unambiguous names to these groups while specifying the occurrence of each separate group or item within a group.
In commercial work simplicity of input and clarity of output are of the greatest importance and particular attention was paid to these. For example, input information from cards can be described in terms of the position of a field on the card while printing is described in columns and rows.
The system has been criticised because of the slow compilation speed.
It was unfortunate that the ingenious engineering ideas in ORION 1 proved unsatisfactory in practice because the use of NEBULA never progressed beyond ORION 1 and 2 whereas it might have become much more widely used and an important Ferranti contribution.
The Sirius computer grew out of "NEWT" which was the test bed for the neuron circuits used on ORION. It was decided that in its original form this test bed needed redesign, and the engineers preferred to do this rather than make a transistorised Pegasus. In view of what happened to ORION, the Sales Department should have pressed more strongly for the transistorised machine. Sirius was also used to exploit another of Gordon Scarrott's developments, the torsional long delay lines in which pulses were translated into twists of the wires instead of into sound waves.
SIRIUS is a decimal machine, in its basic form it had a store of 20 nickel delay lines each with 50 locations making a total capacity of 1,000 locations, each of which holds ten decimal digits. Including 0, 10 numbers could be expressed in the SIRIUS code.
When the ten decimal digits are to be interpreted as an instruction, the first six specify an address, the next two the function to be performed, the next one specifies the accumulator to be used and the final one relates to the modification facilities in the code.
Unfortunately Sirius had to compete against well established computers like the Elliott 803 and did not have any marked superiority over them.
It did however prove to be a very popular teaching machine. It could be operated manually and taken through a programme instruction manually or automatically at slow speed, with a decimal display as required of the contents of the store. The machine code and autocode were very popular. This close relationship between the computer and the operator was one of the features which made SIRIUS excellent for students and for outside users. At one time the Blackburn College of Technology and Design had 7 SIRIUS computers, more than half the total sold on the U.K. market. Several of these have been passed to schools and are still in operation in 1973.
The SIRIUS computers were mostly used for mixed technical and commercial work, both inside the owners' organisations and servicing outside customers.
Out of the 10 delivered to U.K. customers between 1960 and 1963, six were still in use in 1971 though most had been removed from their original location.
The Australian venture was the result of a number of influences which came together, but it is difficult now to say in what order they arose. Trevor Pearcy had made one of the earliest computers in Melbourne University, John Bennett was from Queensland and had returned to Australia to work on SILLIAC, the copy of the Illinois University machine at Sydney; the Ministry of Defence was planning a large computer installation at Canberra and the Australian Government sent Barry Pridmore to the U.S.A. and Europe to study which computer they should buy for unifying the Government administrative statistics; the Commonwealth Scientific and Industrial Research Organisation were preparing to install a large computer. Ferranti were well placed to take advantage of the movement. They had close connection with the Wilson Transformer Company at Melbourne and shortly were to buy a substantial share in that Company while at Woomera they had a useful team which it should be possible to build on.
The Australian market was wide open for the time-sharing kind of computers we were building and a conference was to be held at Sydney in 1960 which would be an opportunity to introduce our new ideas.
So when E.M.I. wanted me to join them it seemed sensible to suggest that I would prefer the two companies to work together and that a joint exercise in Australia would probably be the best way of trying out a joint organisation, far enough away not to disturb the separate activities in Europe. We had not then reached the point of trying to sell Atlas in America and I do not think that in the first instance we expected to be able to sell so big a computer in Australia, though it soon became evident that they would think in ambitious terms.
Talks with Jack Wilson took place in the autumn of 1959 and there was a possibility of support from a carpet manufacturer friend of his. It was agreed that George Felton and I should go to the Sydney Conference where George could read a paper on the Orion Computer. We went at the end of May 1960 and saw a number of people including Insurance Companies, Banks, Broken Hill Proprietry, the Snowy Mountains Scheme and various organisations in Canberra. We left Arnott and Foden to continue the investigations.
During the conference we met Barry de Ferranti who was then working for I.B.M. in Sydney. Afterwards he telephoned Sir Vincent and said he would like to join the Ferranti Co. There were obvious advantages in having one of the family and it was decided that as a member of the family he should head any Australian venture. Unfortunately this resulted in our later losing Bob Arnott.
At a meeting in West Gorton in August it was agreed that a Sales Office should be opened in Melbourne and that a Sirius Computer should be sent out. It was further intended that as soon as possible an Orion should be established in Melbourne and negotiations began to try to arrange this in collaboration with a firm of Accountants, Wilson Bishop and Henderson. It was also agreed that Arnott and Nicol should go to start up the exercise, Arnott having the title of Deputy Australian Sales Manager.
Bob Arnott was an electronic engineer who had worked at Bracknell and, in particular, as designer of the Apollo Computer at Prestwick for Air Traffic Control. His family were very well known in Australia as leading manufacturers of biscuits. Bob had also been a member of the Australian ski team at the Olympic Games at Helsinki and was still the Australian representative on the International Committee. He was also a skilled pianist and extremely popular wherever he went. It was understandable, but a loss to us, when he decided he could not continue in a junior position.
Barry de Ferranti spent several months in the U.K. in 1961 planning the arrangements and recruiting staff. By September of that year we were able to report that Monash University would probably want a SIRIUS or a PEGASUS. The Vice Chancellor of Monash was Matheson, who we had known as Professor of Mechanical Engineering at Manchester and a good friend of Ferranti. He had started the University; when I first visited him it was only a board in a field plus the house which had been bought for the Vice Chancellor.
There was also some demand for medium sized computers and prospects for ORION were good, if it had worked they would have been very good. I.C.I, told us in London that they wanted a computer for the ICIANZ Company and A J Young favoured Ferranti because he felt they would continue in Australia.
By June 1962 we knew that CSIRO were prepared to spend million, possibly on ATLAS; Queensland University were in the market with about £120,000 to spend.
The Computing Services based on SIRIUS made a slow start because of the unreliability of the computer. It seemed that information about certain weaknesses in the manufacture which had been corrected in the U.K. had not been advised to Australia and this caused some feeling on the other side of the world, but quite soon we were being told that the service was the best in Australia.
The long delay in getting the Orion computer to work was a great disappointment and it was too late to establish a market for Pegasus because it was not transistorised. The concentration therefore had to be on SIRIUS and ATLAS and it was because Australia became the most likely place to sell ATLAS that the exercise seemed worthwhile. Had we been able, as seemed very likely, to pull off the sale of two ATLASES it would have been relatively easy to set up a powerful, viable unit in Australia.
This would also of course have made an entirely different appearance in the figures. In the early days I had kept separate accounts for the selling costs and the computing service and had for my own guidance looked on the combined activity as belonging to one account. This had the advantage of insuring that the selling charge against the department should always be kept below ten per cent of input. This idea was objectionable to management and could not be formalised. In Australia however an account of this kind was demanded by management. Inevitably costs had to exceed income in the early stages and in Sales we were concerned that this excess could be borne within the total cost of the Sales Department. Pending the arrival of a suitable Orion type computer the main purpose became to sell ATLAS for which the factory in Manchester needed orders. There were obvious differences between the accountants who regarded the deficit as a "loss" and the Sales Department who regarded it as part of a Sales Cost. It had never been expected that a Sales activity could be entirely supported by the surplus on a computing service, as later seemed to be implied by the accountants.
When the business was sold to I.C.T. the computing service income at a rate of about £20,000 was probably about self-supporting and could have provided a useful base for sales when suitable computers were available. Unfortunately the CSIRO orders were lost to C.D.C. who were able to demonstrate a working computer while the first ATLAS was unfinished, and Barry de Ferranti still had only SIRIUS to sell. I.C.T., with their 1200 and 1301 computers, were able to build up a substantial preparation for the 1901 when this came along.
In January 1963 Computer Consultants gave the following figures for the value of British Computers installed in Britain:
£1,000 Ferranti 6,620 I.C.T. 6,614 English Electric 3,565 Elliott Bros. 3,178 N.C.R. 3,060 LEO 1,740 A.E.I. 1,140 S.T.C. 540 Various 570 TOTAL 27,027
Since Elliotts and N.C.R. were working closely together, there were three groups with approximately equal home figures with English Electric having about half the value and the others behind. The I.C.T. figures included, at this date, the E.M.I, ones which perhaps accounted for one quarter and imported machines probably for about a further quarter of the I.C.T. total. Ferranti computers were home made but included a substantial proportion of imported peripherals. By this date I.B.M. had less machines installed here than I.C.T. but the value was probably somewhat greater.
The companies above - and I.B.M. - were the Ferranti competitors through the twelve years.
When we started selling there was no real competition but no easy sales. The enemy was the inertia of the customers who were studying what was happening before they could move. Our immediate competitor was English Electric with their DEUCE machine. By the end of 1956 they had delivered eight while we had delivered only three Pegasus, though we had dispatched eight of the Mark 1 and Mark 1* and so we were ahead in value. Pegasus deliveries overhauled DEUCE in the following year and then kept somewhat ahead.
Elliott Bros, were, we felt, the most serious competitors to Pegasus and in the early days we were competing for the same customers through most of the 1950's. They became powerful competitors when they introduced the 802 and 803 machines in 1958/59.
Before the creation of I.C.T. the Powers P.C.C. took a considerable number of orders and though the computer was a failure it did of course injure its competitors and perhaps Ferranti in particular. The B.T.M. 1200 series, though regarded as sophisticated calculators rather than computers also satisfied a substantial part of the market.
We were able to keep I.B.M. out of the lower end of the market. Their 650 took a few orders against Pegasus and other small computers until about 1960 they introduced the 1400 series, and particularly the version which had a large random access memory. Then in the 1960's they began to deliver the 7000 series in this country to which our answer was ORION and ATLAS. Had ORION been a success we should have held our position.
It was with the coming of the E.M.I. 1100, which was transistorised that we were forced to launch ORION. English Electric were also selling their KDF9 and KDP10 computers and Lyons the LEO III at the top end of the market. The difficulties with ORION, the lack of a successor to Pegasus other than the light-weight SIRIUS made it impossible to stop I.B.M. The principal competitor to SIRIUS was the Elliott 803 which followed their earlier core-store computers, but the I.B.M. 1620 also competed.
ORION was against competition from such machines as the I.B.M. 7070, LEO's KDP10, I.C.T. 1300 and I.B.M. 1401. It was in the market which this last machine was capturing that we ought to have been able to sell the smaller ORION's and, when it was clear there were to be no ORIONs small and cheap enough, the PEGASUS successors.
There were a number of the I.B.M. 1400 series which would have been due for replacement about the time that the PEGASUS successor was coming along and this looked like a market to be cultivated. The question of I.B.M. compatibility would have been raised and would have had to be solved, but this was becoming quickly more and more seriously necessary.
ATLAS competition was from the largest I.B.M. computers and, most seriously, from C.D.C., but in the end I.C.L. took a decision to stop this computer in favour of the larger 1900's.
The competition situation among the British machines continued far too long, putting the Government in a difficult position as it tried to decide which horse to back. They were in the end compelled to bring the engineering companies together with the unified punched card companies. The long delay had, however, made unity more difficult to achieve so that the new I.C.L. was only viable with a good deal of Government support which it has continued to need ever since.
ORION was designed as a time sharing computer to be shared among a number of programmes in such a way that they would appear to run simultaneously. The Organisation and Monitor Programme ensured that the jobs could not interfere with one another and kept the operator fully informed of the progress of the computation.
For each job the computer was informed of the programme name and the data to be used. This information could be given by a piece of tape or a small pack of cards. OMP recorded the necessary identification details and determines the peripherals to be allocated to it. This could only be altered by OMP which could therefore ensure that each programme used only its own peripherals.
In the core store special registers held the first and last addresses that the current job might use and these were compared with each instruction as it was obeyed so that the hardware could prevent any attempt to violate the reserved space. Other routines in the system dealt with the abolishing of finished jobs and the release of their store space and peripherals.
The monitoring part of OMP arranged for messages to be printed out to tell the operator what was happening, for example, information about the state of peripheral or a violation of a reservation in the core store. When a job was finished details were printed out of the amount of machine-time used etc. A result of this method of communication with the system was that all operator actions and other important incidents were recorded on a Flexowriter output.
In a time sharing system the timing of Jobs is particularly difficult. The system arranged that every minute the time of day was printed and punched on the Flexowriter. The starting and stopping times of each job was recorded so the elapsed time could easily be determined, but, since this may have little relation to the time spent by the central computer a timer register was introduced. This is zeroed when a job is started and the amount of time the "mill" of the computer was used on the job was added to a partial total belonging to the particular job. This timing system was used to provide an additional check. The partial total belonging to the job was set to a figure which was the negative of the time initially asked for for the job so that if this figure turned positive an interruption could call for the OMP programme to examine the state of the job.
The first jobs done on the computer came from the University and its connections. Cyril Gradwell, one of the first Ferranti programmers, did a calculation for the Shirley Institute to show the dynamics of a cotton thread when passing through the air during ring and cap spinning. The first paid job after we opened the office in Moston was a problem of costing in actuarial work. There followed a stream of calculations in the design of aircraft, frame structures, the costing of paint production for I.C.I., computation of index numbers, and so on. We did an experiment, using the computer to make the least squares adjustments of traverse calculations for the Ordnance Survey and one of unusual interest which went on for a long time was to find for Pilkingtons a way of minimising the waste which arises when plate glass has to be cut to remove portions with flaws. We experimented with other commercial procedures and found the calculation of wages more difficult than we had expected; John Bennett introduced the Transformer Department to computers and over the years a lot of work was to be done and some computers sold in this field. A selection of the early jobs is listed, with brief descriptions, at the end of this section.
In the early days some jobs were done to demonstrate the facilities of the computer and others were for possible buyers. We were able to charge part of the cost of these demonstrations in most cases because serious potential purchasers realised they were getting valuable information and often getting real jobs done; the willingness to pay was an assurance that the demonstration was of interest to management and not just to an enthusiastic technician who did not have any financial powers. We were often told that the punched card firms did demonstrations free but we could point out that these did not involve the programming effort in which we were involved.
It soon became clear that although programming was burdensome, getting the facts was more so and the best way of working was to get the man who understood the problem to come and work with a programmer who understood the computer, otherwise it was nearly impossible to ensure that the programmers got accurate information with all the detail they needed.
The association between the programmers and the customers' staff undoubtedly paid off in the early stages but we made the mistake of neglecting the "glossy brochure" approach which became important later.
This was illustrated in the case of A.V. Roe. Their first computer they bought because they had worked on the Manchester Computer and convinced themselves of its use. When they came to want one for production control I told the Managing Director we would very much like to quote and could we look into the requirements of the job? His response was startling: that was the trouble with the English Companies; they all wanted to look at the job whereas I.B.M. gave him a booklet which showed him how to do his production control. He waived a slim and colourful pamphlet and installed I.B.M. equipment. But of course his approach could only convince when it was possible to refer to a large number of successes in this or very similar fields, and perhaps few managing directors are so ignorant of their organisations.
To do real jobs against time was a necessary but severe test of the computer which caused much tension between the engineers and the programmers. No doubt the programmers expected too much of an early computer and the pressure by the sales staff to have it working for the benefit of possible buyers added further pressure on maintenance staff who were doing a very difficult job. Fortunately at the working level on the computer relationships were good. The real trouble was in the difficulty of telling when a fault lay with the computer and when in the programme.
Techniques were developed for breaking jobs down into sections small enough to make fault finding easier. This was one of the areas in which we missed the convenience of punched card equipment but the real solution came when the method of building with logical packages separated the two types of problems.
It came as a surprise to find how much of the trouble came from mechanical equipment which had been in use and perfected over long periods. Power supplies gave a good deal of trouble and so did electro-mechanical ancillaries. There was a fairly serious suggestion that we should install a large battery system to provide the necessary stable dc voltage for the Mark 1 computer in the University.
Users of the computer were remarkably tolerant even when they had to travel long distances and sometimes wait many hours and perhaps go away without doing much useful work. The University and Ferranti staff sometimes worked very long hours to get the necessary minutes of production time, but it was impossible to continue to exploit enthusiasm even when, as one of our girl programmers put it, the job was like being paid to have fun. There came a Monday morning on which I learned that one of the girl programmers had started work at 9 a.m. in the factory on the previous Friday, gone to the University in the afternoon and, to finish the job, had continued until mid-day on Saturday. More important, from a commercial point of view, was that travel to Manchester was still slow and from London the return journey could not be done in the day. To get senior people to devote one and a half to two days to see the computer was difficult; but they needed to touch a computer before they could really accept its existence. We learned the hard way that sales are directly related to the number of contacts with buyers and these contacts were too few; the local Lancashire firms were after customers of the University rather than Ferranti and, anyhow, they were too few. Most of the businesses and research establishments using the early computers were located in the South.
By the end of 1952 we had enough evidence to show we ought to be able to justify putting a computer in London and Sir Vincent agreed to discuss it with Messrs Bass, Grundy, Robson and Swann. After three hours in which the Chairman, Bass and Robson put up all the arguments as to why we ought to be able to do the job in Manchester the meeting broke up. The following morning Sir Vincent called me in "to discuss the problem quietly". Part of the argument had been whether N.R.D.C. should join us in the venture and after emphasising the dangers the Chairman said "he thought we should do it and that we should do it ourselves". He had clearly seen the need to get as many people as possible attached to our computer and this had to be in London. (In the last six months before I.C.T. took over earnings in London were £70,000 compared with £5,600 in Manchester, £2,750 in Edinburgh and over £8,000 in Australia).
The problem of finding premises was explored when Hugh Ross joined us and, shortly after, Chris Wilson. They scoured the possible areas in the West End with little success; then one day I consulted an estate agent in the Baker Street area who immediately said we ought to take one of the old houses on the north side of Oxford Street because the price per square foot was half as much as on the other side. This produced No 21 Portland Place. The house was built about 1780, a part of the development launched by the Adam brothers to build a "street of palaces".
This building, with four floors and a basement, proved remarkably - and surprisingly - suitable for our purpose. The floor had to be strengthened to take the computer, but the large Georgian reception rooms made a lecture room and a conference room on the ground floor with the computer in a first-floor drawing room with a painted ceiling. It proved a memorable place for any of the early students or programming.
We were fortunate that by the time we were able to move in - at the beginning of 1954 - we could see we did not need to have a Mark 1* computer and we prepared for a Pegasus, installation of which began in 1955. The house which we expected would have accommodation for about 20, held at one time about 60 staff.
In our original memorandum to show the scheme should be viable, we allowed for 10 programmers, 5 engineers and 10 other staff, estimated to cost £45,000 per annum (rent and rates were only £5,000 for nearly 10,000 sq.ft. virtually all usable space). Income from a limited number of customers with whom we were in touch we could estimate as £39,000 per annum, only using half the machine's time. A copy of the guesstimate is shown at the end of this section.
One of the arguments for the computing service machine as a selling aid was that programmers would like to continue to use a machine they knew and when it became clear that Pegasus was to be a programmers' computer our confidence in this approach increased. Some of the customers and potential customers who had learned their programming at Manchester transferred easily and eagerly to this new computer and their work helped to build up the library of programmes. It was soon possible to run programming courses to train customers' staffs and others.
We should have exploited the interest in the novel development more by giving more courses. We could perhaps have built up a business in systems consultancy, but we were held back by reluctance to expand staff and doubt whether the world would accept and pay for consultancy from a manufacturer. LEO Computers were approaching the problem this way but most potential computer customers had been brought up by the punched card firms who did all the necessary investigation free. We should have been able to build up the computing service faster, but capital expenditure was unpopular and only slowly did we find that the market was big both BTM and Powers had been emphatic that a computation service was a necessary, but costly, sales aid. By 1962 we had over 150 customers using Pegasus and Sirius computers in the service and it was clear that services based on electronic computers would be a permanent feature for as long as could be foreseen.
There was another hold up. When Welchman set up the research group which was to work with Powers and to help the engineers who were building the Mercury computer in Manchester, he was also given the computing service. This had the admirable result of bringing in Stanley Gill and Alan Bagshaw but produced a somewhat confusing organisation. It was still necessary to have sales programmers for demonstration purposes and these continued to use the greater share of computer time while Bagshaw worked up new business. Happily the individuals worked well together and the computing service staff on Pegasus and the sales programmers made almost a single team. It did mean however that for several years the figures relating to this activity were confusing and often meaningless.
The arrangement had one effect on the figures which probably affected the argument which later developed between N.R.D.C. and Ferranti when the whole Pegasus exercise with N.R.D.C. went into the red. The cost of the selling staff was chargeable to N.R.D.C. as part of the costs of the Mark 1* and ten Pegasus computers we were selling on their behalf. But any money earned by these sales programmers was credited to Welchman's computing service account. A number of his staff were employed mainly on the Mercury computer and in this way Mercury may have been relieved of costs at the expense of Pegasus, which meant N.R.D.C.
As soon as Pegasus became a reality we concentrated on building up the computing service in London. We did not expect that Ferranti would be able to do much work on the Mercury in Manchester because the kind of work it would be doing would be pre-empted by the University. Pegasus started working fully in June 1956 and the following tables show the build-up of the service and something of the split between different kinds of activities.
The Pegasus Computers were contributing more than three-quarters of the earnings which were as follows for the months of April and May, 1963:
£ Pegasus 9,192 Sirius 2,079 Atlas 215 Manchester (mostly Pegasus) 170 Edinburgh (mostly Pegasus) 447 Mercury 60 TOTAL 12,164
This work was divided nearly equally between scientific and commercial work as follows (except for £33 worth about which it seems we could not make up our minds)
Scientific 6,434 Commercial 5,697 TOTAL 12,131
In the early years the main use of the service was for training programmers and for preparing their programmes. Thus in the year 1957/58 two-thirds of the computer earnings came from firms who had bought computers and most of the rest from firms we classed as potential customers, only about £1,000 was from small users who were unlikely to be buyers. As computers were delivered the use by owners fell off quickly and most of the income was supplied by potential customers. The figures for 1953-59 and 1960-61 were also depressed because the service had to give up the magnetic tape attachments. The factory were unable to deliver equipment ordered by the Transformer Department and it was decided that our units should be sent to them and replaced by the next set available. This ended in a nonsensical - but costly - arrangement. By the time the London units had been dismantled and refurbished their replacements were available and the Transformer Department got the secondhand set several days later than Portland Place received the new ones. By the first half of 1960/61 only one tenth of the income came from computer owners (£1,500) and small users, but when the sales of computers rose rapidly in that year the new owners' need for computer time increased, giving good figures for 1960/61. Small users began to rise and reached £8,000 in the first quarter of 1961/62. By this time we had a small staff devoted to selling the computing service as well as encouraging the computer salesmen to sell the service as a lead-in to selling computers.
Up to the time when I.C.T. bought the Department we were still mainly dependent on Pegasus, but Orion was beginning to work and we estimated an output for 1963/64 of £428,000. We can now see from the performance of SCICON (formerly CEIR), who developed the kind of activity we began, that a very useful business could have developed. SCICON profits are I believe of the order of £1M. per annum.
The various jobs produced a steady stream of pamphlets - C.S. Lists - describing programmes and their applications. These began as internal memoranda but became very popular with customers and made first-rate advertising material. We ought however to have produced more illustrated and coloured pamphlets and could have prepared and sold small text books on the way computers had been and might be employed.
It was anticipated, and experience has confirmed, that it would be necessary for a computer to be available for demonstration in London as Manchester is too far away.
Several properties have been investigated and one - at 21 Portland Place - may be taken as reasonably typical of the kind of premises needed. The house has a floor area of 5,000 sq.ft. a room large enough to take a computer and sufficient accommodation for about 20 persons, room to give lectures or for training courses for people who want to use computers.
Outgoings are provisionally estimated as approximately:-
£ Rent, rates, etc 5,000 Hire of computer (for 8 years therafter £1,100) 9,500 Salaries of say 10 programmers 10,000 Salaries of 5 engineers 4,500 Salaries of 10 other staff 10,000 Plus other expenses and contingencies 16,000 TOTAL 45,000
A total of about £45,000 seemed reasonable
A survey of customers who have indicated that they want service work done on the computer gives the following rough estimate of earnings:-
R.A.E. 8,000 Ordnance survey (figure to follow) Ferranti factories 2,000 Aircraft companies 13,500 Board of Trade 2,000 British Electricity Authority 1,000 Hire of machine to bodies Universities etc. 7,500 Charges for Programmers Time 5,000 Notional amount for charge against selling costs of computers 5,000 TOTAL 44,000
Without going beyond a limited number of customers with whom we are at present in touch, expenses should be covered. The jobs considered above should not occupy the machine for half of its time so there is scope for comfortable expansion and very fair prospects of substantial profits.
In the second year, if not before, it should not be difficult to expand the work considerably. The principal reason for being cautious about the first year is that programming of new jobs takes a long time and it may be wise not to assume too much expansion in the first year.
The figures relate to a notional year beginning when the machine is erected in London.
Date: About March 1953.
|Statistics of Input, Output and Unexecuted Orders|
|Up to 1954||est 1,000||est 800|
1. Up to 1960-61 Digital Systems Department was part of the Computer Department. The figures in brackets are the estimated D.S.D. figures included in the totals.
2. The figures for unexecuted orders do not precisely tie up with the Input/Out totals for individual years.
|1.||Manchester University||1951||Mathematical research|
|2.||Toronto University||1952||Mathematical research|
|3.||Ministry of Supply||1953||Classified|
|4.||Royal Dutch/Shell Laboratories, Amsterdam||1954||Oil Refining Studies|
|5.||National Institute for Applications of Mathematics, Rome||1955||Research work|
|6.||Atomic Weapons Research Establishemnt, Aldermaston||1954||Research work|
|7.||M.O.S. Fort Halstead||1955||Research work|
|8.||A.V.Roe & Co. Ltd., Manchester||1954||Aircraft design calculations|
|9.||Armstrong Siddeley Motors Ltd., Coventry||1957||Research work|
|1.||Ferranti Ltd, Portland Place, London||Mar 56||General computing service work||General computing service work|
|2.||Hawker Aircraft Co. Ltd.,Kingston-upon-Thames||Oct 56||Aviation design calculations|
|3.||Admiralty Research Laboratory, Teddington||Feb 57||Research work|
|4.||Armstrong Whitworth Aircraft Ltd.,Coventry||Nov 56||Aviation design calculations, analysis work||Payroll|
|5.||Royal Aircraft Establishment, Farnborough||May 57||Research calculations|
|6.||Vickers-Armstrong (Aircraft) Ltd., Weybridge||May 57||Aviation design calculations|
|7.||I.C.I.Ltd. Dyestuffs Division, Manchester||Dec 57||Research work||Sales analysis and forecasting, stock control|
|8.||N.R.D.C,, Northampton Polytechnic, London||June 57||Research and training|
|9.||De Haviland Aircraft Co. Ltd., Hatfield||Aug 57||Research work||Payroll, budgeting|
|10.||British Thomson-Houston Co.Ltd., Rugby||Aug 57||Turbine design work||Costing|
|11.||British Iron & Steel Research Assoc., London||Nov 57||Operational research work|
|12.||Leeds University||Oct 57||Research and service work||University registration work|
|13.||Durham University||Oct 57||Research and service work||University registration work|
|14.||Southampton University||Mar 58||Research and service work|
|15.||Babcox & Wilson Ltd., London||Jan 58||Research work||Stock control and management accounting work|
|16.||The United Steel Co., Sheffield||Jan 58||Operational Research work|
|17.||Blackburn Aircraft Ltd., Brough||Mar 58||Research work||Production control investigations|
|18.||Svenska Flygmotor A/B, Trollhatten, Sweden||Jun 58||Research work||Production control|
|19.||M.O.S., Military Survey, London||Aug 59||Survey calculations|
|20.||Stuttgart University, Germany||Jun 58||Research and service work|
|21.||Ferranti Ltd., Hollinwood, Lancs||Aug 59||Transformer and technical calculations||Production control, wages, service work|
|22.||C.A.Parsons & Co Ltd, Newcastle||Jan 59||Transformer design work|
|23.||The Steel Company of Wales Ltd., Port Talbot||Feb 60||Operational research work|
|24.||Ferranti Packard Electric Ltd.,Toronto, Canada||Dec 59||Transformer design work and service work|
|25.||The College of Aeronautics, Cranfield||Sep 60||Aircraft design calculations|
|26.||Aircraft Armament Exp. Est., Boscombe, Devon||Dec 61||Aircraft design calculations||Data analysis|
|27.||Skandia Insurance Co., Stockholm, Sweden||Dec 59||Actuarial work|
|28.||Ferranti Ltd., Newman St, London||Aug 60||General computing service work||General computing service work|
|29.||Bruce Peeebles & Co, Edinburgh||Aug 60||Transformer design work||Production control investigations|
|30.||D.S.I.R., Road Research Laboratory, Harmondsworth||Jan 61||Research calculations||Accident record analysis|
|31.||London and Manchester Ass. Co. Ltd., London||Oct 60||All ordinary branch insurance work; investment work|
|32.||Shell Refining Co. Ltd, Stanlow||Feb 61||Refinery technology work||Payroll and stock control|
|33.||Shell research Ltd., Thornton||Feb 61||Technologiacl work|
|34.||Vickers-Armstrong (Aircraft) Ltd., Weybridge||Jul 61||Aircraft design calculations|
|35.||De Havilland Propellers Ltd., Stevenage||Aug 61||Aircraft design calculations|
|36.||Martins Bank Ltd., Liverpool||Apr 61||Current account book-keeping|
|37.||The Scottish Widows Fund & Standard Life Assoc. Soc., Edinburgh||Apr 62||Pension scheme updating, ordinary branch evalustions|
|38.||Westminster Bank Ltd., London||1962||Current account book-keeping|
|1.||Norwegian Defence Research Establishment, Kjeller||Aug 57||Atomic energy work|
|2.||Manchester University||Oct 57||Research and services work|
|3.||French Atomic Energy Authority, Saclay||Nov 57||Atomic energy work|
|4.||United Kingdom Atomic Energy Authority, Harwell||Feb 58||Atomic energy work|
|5.||RAF Meteorological Office, Dunstable||Sep 58||Weather forecasting|
|6.||Council for European Nuclear Research, Geneva||Jun 58||Atomic energy work|
|7.||London University||Oct 58||Research and service work|
|8.||United Kingdom Atomic Energy Authority, Risley||Oct 58||Atomic energy work|
|9.||Oxford University||Nov 58||Research and service work|
|10.||Shell International Petroleum Co. Ltd, London||Jan 59||Linear programming||Sales analysis|
|11.||Royal Aircraft Establishment, Farnborough||Mar 59||Aircraft calculations|
|12.||I.C.I. Ltd., Central Instruments Division, Reading||Jun 59||Chemical process analysis|
|13.||Swedish Atomic Energy Authority, Stockholm||Jul 59||Atomic energy work|
|14.||Belgian Atomic Energy Authority, Mol.||Sep 59||Atomic energy work|
|15.||The General Electric Co. Ltd., Erith||Dec 59||Atomic energy work and machine design|
|16.||Metropolitan-Vickers Electrical Co. Ltd, (A.E.I.), Manchester||Oct 60||Transformer design|
|17.||United Kingdom Atomic Energy Authority, Winfrith Heath||Jun 60||Atomic energy work|
|18.||Buenos Aires University||Sep 60||Atomic energy work|
|19.||British Petroleum Co. Ltd., London||May 61||Linear programming||Sales analysis|
|1.||AB Datacentralen (Trygg a Fylgia Insurance Companies), Stockholm||Apr 59||Ordinary and motor insurance, policy updating, premium notices, etc|
|2.||South African Mutual Life Ass. Soc., Cape Town||Dec 59||Insurance policy updating, premium notices, etc|
|1.||Ferranti Ltd., Manchester||1963||General computing services work||General computing services work|
|2.||Ferranti Ltd., Newman St, London||Mar 63||General computing services work||General computing services work|
|3.||AB Turitz & Co., Gothenberg, Sweden||mar 63||Stock control and sales forecasting|
|4.||National Institute for Research into Nuclear Science, Harwell||Aug 63||Atomic Energy work|
|5.||General Electric Co.,Ltd, Birmingham||cancelled||Research calculations||Production and stock control and payroll|
|6.||The Prudential Insurance Co. Ltd. (Orion 2 replaced Orion 1)||Sep 64||Ordinary branch insurance work|
|7.||Norwich Union Life Insurance Society, Norwich||Jan 64||Policy updating, premium notices, valuations|
|8.||Rothampstead Experimental Station, Harpenden||Dec 63||Statistical work|
|9.||National Provincial Bank Ltd., London||Dec 63||Current account book-keeping|
|10.||Beecham Group Ltd., Brentford||May 64||Stock control, invoicing customer accounts, sales analysis|
|11.||Metal Box Co.Ltd, Worcester||Mar 64||Payroll, invoicing and statistical work|
|12.||Vickers Armstrongs||Jan 65|
|13.||Cadbury Bros.||Aug 64|
|1.||Ferranti Ltd, Newman St, London||Jun 60||General computing services work||General computing services work|
|2.||Ferranti Ltd, Newman St, London||Feb 61||General computing services work||General computing services work|
|3.||Yarrow &Co. Ltd., Glasgow||Oct 61||Pipe stressing calculations||Technical and commercial work and external service|
|4.||Cement and Concrete Assn, Slough||Feb 62||Frame stressing work|
|5.||Ferranti Ltd., Melbourne, Australia||Nov 61||Service work|
|6.||Kovo, Czechoslovakia||1962||Technical work|
|7.||Imperial Chemical Industries, Melbourne, Australia||Feb 62||Scientific and Technical|
|8.||Ferranti Ltd., Melbourne, Australia||Feb 62||Service work|
|9.||Builders Copper Tube||1962||Invoicing etc|
|10.||Monash University, Melbourne, Australia||Jul 62||Adminstrative and research work|
|11.||Davy Ashmore||Sep 62||Design of still making equipment, research and performance analysis|
|12.||Herriot Watt University||May 62||Teaching|
|13.||Admiralty, Bath||Mar 63||Design and construction calculations for naval vessels|
|14.||British Railways||Apr 63||Technical work||Bonus calculations|
|15.||Trumpy y Sirvent, Madrid||1963||Service centre||Service work|
|1.||Manchester University||1963||Research and service work||University registration work|
|2.||London University||1963||General computing service for the university and service bureau for outside organisations|
|Sales costs as percentage of:|