Literature: Reports

Jump To Main Content

Jump Over Banner

Home

ACLLiteratureReports

Jump Over Left Menu

The Atlas Computer Laboratory

Jack Howlett

1993

IEE Committee S7 Warwick Meeting, July 1993
IEE

Background

This is about something that started in the late 1950's, only a little over thirty years ago; very recent by the standards of much of the history of electrical engineering, but a very long time ago in the history of the electronic digital computer. But first let me go back some 10 years earlier. I joined the newly-created Atomic Energy Research Establishment (AERE), Harwell, in 1948, to be responsible for computing in the Theoretical Physics Division: the electronic computer scarcely existed then, only as a few tremendous adventures in a small number of University and government laboratories here and in the USA, most notably here in the universities of Manchester and Cambridge and the National Physical Laboratory. Most certainly there were no machines to be bought. Commercial computing on any large scale (data processing, really) was done on punched card machinery; scientific computing, as at Harwell, on hand-operated desk calculators.

It's well to remember that the first properly-engineered computer to be built and sold as a commercial product was the Ferranti Mark 1, in 1951. But it has always seemed to me that the computer industry really started around 1955 with IBM 650, a valve machine comparable with and perhaps a less elegant design that the comtemporaneous Ferranti Pegasus; Ferranti sold 40 of these and were very pleased with that success, but IBM sold about 2000 of the 650 and immediately began to dominate the industry.

By the late 1950's the computer had become thoroughly established, but the scene was utterly different from to-day. Valve technology had been replaced by solid state, but integrated circuitry was only just dawning and all computers were built from discrete components. The name "computer" was almost synonymous with "mainframe", and mainframes were all very large, very power-hungry, very fussy about their environment and very expensive. Data transmission over telecommunication links scarcely existed, networks not at all, and the personal computer hadn't even been thought of; so far as I can recall, neither had the term "Information Technology". The most powerful machine on the market was the IBM 704, with its almost incredibly large random access memory of 32k words (128Kb) on ferrite cores; this was backed by magnetic tapes, disks being not yet in production. The most powerful British machine was the Ferranti Mercury, comparable to the 704 in basic speeds but on an altogether smaller scale and with a core memory of 1k words, equivalent to about 5 Kb: this was the most powerful machine installed in any British university. For the industrial comparison again, Ferranti sold about 20 Mercurys, IBM about 300 704s.

Harwell

AERE Harwell was created to do the underlying research and provide the numerical data needed to support the design and building of nuclear power-generating reactors. Under it's first Director, Sir John Cockcroft, it was a real power-house of scientific talent and a wonderful place to work. I must emphasise that Harwell was not a weapons laboratory - Aldermaston was set up for that; security restrictions were imposed on almost everything to do with nuclear (then called atomic) energy at that time, but Cockcroft was always battling for greater openness - with much success, in fact.

As I have said, I went to Harwell in 1948 to be responsible for computing for the Theoretical Physics Division; this, by the way, was at the invitation of the then head of that Division, Dr. Klaus Fuchs, of whom much was heard later. It's an interesting comment on the attitude to computing at the time that this very large and superbly equipped laboratory was set up without any provision for a station-wide computing service. During my first decade there we at first did everything on commercial desk machines - who remembers those classic names Brunsviga, Facit, Friden, Marchant, Monroe? - and quite soon built up a powerful punched card installation which included two of the very sophisticated BTM 555 electronic calculators, which were almost computers. In 1952 we had a computer of our own, designed and built by the Electronics Division, which used dekatron (10 cathode) tubes for arithmetical operations and relays for logic and switching - very slow but immensely reliable: this can still be seen in the city science museum in Birmingham; and we began to take an increasingly active interest in the computer developments at Cambridge and Manchester, later doing real-life reactor calculations on the Ferranti Mark 1* installed at Aldermaston, where we were able to get time on the night shift. In 1958 we installed a Ferranti "Mercury" computer, but for most of the reactor design calculations we needed to use the IBM 704 which then had been installed at Aldermaston.

Future Needs: The Harwell Discussions 1957-58

About this time my colleagues and I at Harwell began to feel concerned about the rate at which Britain was falling behind America in the design and manufacture of computers. I like to think that we weren't so naive as to imagine that Britain could actually rival the USA, with the latter's so much greater home market and industrial capacity; certainly we felt that we were in danger of being completely swamped and driven out of what was clearly going to be a very important market. We could see the need in reactor design for much more powerful computers than were then available from many manufacturer, here or in the USA, and we felt that this presented an opportunity that Britain should seize. We were spurred on by the news that IBM had embarked on the design of what then seemed almost the ultimate in computers, the famous Stretch.

This informal group, consisting of myself K.W Morton (now Professor of Numerical Analysis at Oxford) and E.H. Cooke-Yarborough, head of Electronics Division at Harwell, together with Johm Corner, responsible for computing at Aldermaston, arrived at a reasonably specific idea of the scale of the computer needed for reactor design; in present-day terms the processing speed needed was about 1 megaFLOP (1 million floating point operations per second) - which doesn't sound much now, but represented a real challenge at the time, a "Supercomputer" in fact: Mercury, quite a fast machine, was about 3 kiloFLOPs. We put our views and conclusions to Cockroft, who responded enthusiastically and arranged two informal but very intense discussion meetings of the dozen best informed people in the country. The need for a British project aimed at producing a machine of the kind of performance we had suggested was accepted almost immediately and the discussion focused mainly on how this might be achieved.

At the second meeting, in February 1958, Professors F.C. (Freddy) Williams and T. (Tom) Kilburn described the new machine they were designing at Manchester University; after critical discussions of this and of the other possible routes to the goal it was decided that the Manchester project should be investigated in detail and, if the outcome warranted this, the Atomic Energy Authority (AEA) be asked to support it. Shortly after this Ferranti let it be known that they would design and manufacture a machine based on the Manchester design if they were guaranteed one order. Cockcroft then gave a group consisting of Cooke-Yarborough, Morton and me the task of investigating the computing needs of the AEA - excluding the weapons work of course - and the possibility of meeting these with a machine of this projected design; and reporting our conclusions with specific proposals as we had formed to the AEA by early 1960.

The Atlas Proposal

Ferranti had always given their machines mythological names - Pegasus, Perseus, Mercury, Orion and others; they dictated that if they were to engineer the Manchester design they would call their machine Atlas. The outcome of the activity I have just described was a paper to the AEA which estimated that a machine of the power of Atlas would satisfy the needs reported from Harwell. Risley (the reactor design organisation in Lancashire) and Culham (the fusion research laboratory in Oxfordshire) and still have spare capacity. It made the following proposals:

We regarded this last requirement as very important: university research workers were continually - and not unreasonably, at the time - complaining that they were heavily outclassed in computing resources by their American rivals, and here would be a machine, bought with public money, with time to spare that could be used to help them.

The AEA accepted the paper and agreed in principle to implement our proposals. However, Atlas was going to cost a lot of money, something like £M31/2 at least £M35 in to-day's pounds, and at this cost the project had to be authorised by the then Minister for Science, Lord Hailsham. So the paper with its proposals was submitted to him, and his response was as follows:

An Atlas computer should be ordered, to provide a service to the AEA and to the universities as proposed.

It should not be owned and operated by the AEA; instead, the new National Institute for Research in Nuclear Science (NIRNS), created to build and operate the Rutherford High Energy Laboratory at Chilton, adjacent to Harwell, should be invited to accept the responsibility.

The AEA should pay for the services it received.

I have never been told what were the arguments that led to this decision. It seems likely that one at least was that the AEA was not empowered to give services free of charge to the academic world, whilst NIRNS was set up for just such a purpose; possibly also the fact that Harwell was subject to security restrictions and NIRNS not at all was a consideration. However, it was a most important decision and changed the nature of the enterprise from something like a Harwell scientific Division with an academic appendage to an academically-oriented institution with the AEA as a paying customer. Greatly as I admired Harwell and had enjoyed working there, I have no doubt at all that it was a very sound decision.

The NIRNS Board - whose Chairman, by the way, was that very distinguished ex-Civil Servant, Lord Bridges - accepted the invitation and agreed to build a new laboratory to house the machine and it's staff; this was to be called the Atlas Computing Laboratory and would be situated between Harwell and the Rutherford Laboratory then in course of construction. They set up a Management Committee with a distinguished membership and again a most distinguished Chairman, Lord Penney. The post of Director was advertised, I (not surprisingly) applied and (by no means automatically) got it, and in December 1961 left Harwell to become the single member of the staff of the new laboratory.

Meanwhile, of course, contract negotiations with Ferranti had been going on, conducted by the very experienced AEA contracts organisation; I was very closely involved and found it a most interesting experience. These were completed in the summer of 1961 when Ferranti were given the formal order; in the September they gave a splendid lunch party at the Savoy at which, I recall, I found myself sitting next to Lord Beeching. I recall also looking across the room to Tom Kilburn and exchanging glances which meant that we should never have expected computing to run to such high life.

The Atlas Computer

The technology of Atlas was that of the early 1960s: solid state, discrete components, ferrite core memory backed by magnetic tapes and, originally drums, later by disks when these became available. It was a character-oriented machine, with a word length of 48 bits organised as 8 six-bit characters. It had a "fixed" (read-only) store for basic sub routines, built of small ferrite slugs.

Our installation was as follows:

Main memory
48k works (288 Kchars.), later increased to 64K words (384 Kchars.); paged, 1 Kword pages. Ferrite cores, 2 μsecs cycle time.
Fixed memory
8 Kwords Ferrite slugs, 0.4 μsecs cycle time.
Drums
96 Kwords, replaced first by Bryant and later by Data Products (100 Mchar) disks.
Magnetic tapes
16 Ampex, 1-inch, 2 IBM 729 1/2-inch
Input
2 ICT card readers (600 cards/minute), 1 Ferranti paper tape reader (300 char/sec), 1 Elliott paper tap reader (1000char/sec)
Output
2 Anelex line printers (1000 line/min, 120 char/line), 2 ICT card punches (100 cards/min), 2 Teletype paper punches (110 char/sec)

The raw speeds were:

Floating point addition
1.8 μsec (av.)
Multiplication
5.9 μsec (av.)
Organisational instructions
1.7 μsec (av.)

Giving an average effective speed of about 350 KFLOPs

The total power consumption of the machine, the cooling units and the air-conditioning plant was about 150 KVA

Figure 1: The input/output equipment of the Atlas.

Figure 1: The input/output equipment of the Atlas.
Large View

Atlas Software

Atlas has a good claim to be the first machine with a true operating system, the Atlas Supervisor. This managed all the input/output operations, job scheduling, multi-programming, logging, automatic transfers between main and backing stores and produced very comprehensive operating statistics, which we found of the greatest value in managing the service. Interestingly enough, it was written and developed by a team of about 6 people, headed by David Howarth who now has a Chair at Imperial College - and who carried most of it on his head.

The paged store was a striking innovation, due entirely to the Manchester team. The automatic transfer between this and the backing store, giving what was called the 1-level store, meant that the programmer need not be aware of the physically limited size of the main store. Right from the start the mechanism worked perfectly. The principle was later taken up by other manufacturers and rechristened "virtual memory" - with little or no acknowledgement to Manchester.

Operation and Maintenance

The whole system, hardware and software, had been conceived in terms of batch processing, with everything handled in the machine room. After quite a short time we were running the machine 24 hours a day, on three 8-hour shifts, with 6-8 operators for each shift. We had a "data preparation" staff of about 12, mostly young girls, for card punching (mostly, some paper tape) and doing the considerable amount of clerical work required by the service.

The machine needed regular preventive maintenance, which was done under a contract with ICT. They established a resident team of 6-8 engineers in the laboratory, who had the machine for about 2 hours each day. This scale of maintenance effort was quite normal for a big installation at that time and itself provides a striking comment on the effects of the subsequent advances in technology. There can be very few computer installations now with a resident maintenance team: to-day's integrated-circuitry is many orders of magnitude more reliable than the discrete-component hardware of the 1960's, modular construction makes it simpler and cheaper to replace whole units than to do on-site repairs, and monitoring and fault diagnosis, and indeed most recovery from faults, are done automatically and remotely over telecommunication links.

Setting Up the Laboratory

As I have said, Atlas was ordered from Ferranti in the summer if 1961: one must keep in mind that this was 30 years ago and that it was then seen as a very powerful, very novel and very sophisticated machine. It was going to be physically very large and very demanding on it's environment. The contract required it to be delivered to the new building in April 1964. During 1962 Ferranti sold their computer business to ICT (which in 1968 became a major part of ICL), so almost from the start our involvement was first with ICT and then with ICL - although mainly with the same people.

At the beginning of 1962 I was in the enviable position of having a budget, a piece of ground and the responsibility for planning, creating and then directing a computing service laboratory and putting up a building in which to house it: certainly the opportunity of a lifetime. Even at that distant date there were plenty of large-scale computing services in operation, notably in the big American national laboratories but nothing quite along the lines of what was intended here; it was the greatest good fortune that I and colleagues who later joined me had time to think about how we should like to see the laboratory as a working entity.

The Building

There had to be a building complete and ready to receive the machine in the spring 1964, so the first thing to be done was to get this designed and approved and construction under way. The Ministry if Works, who had built Harwell and were building the Rutherford Laboratory, was then the controlling authority: they assigned us an excellent architect who went to endless trouble to understand the needs we perceived and to produce a design that would meet them. It became clear very soon that the great majority of users of the laboratory's services would be from the universities - in the event, the AEA installed more equipment at Harwell and made very little use of Atlas - and that many would want to work in the laboratory for shorter or longer periods. I was most anxious that they should be made to feel welcome and be at least adequately accommodated, and I wanted the design to take this into account. The floor plans showing what was built are given in Fig. 1, for which a few words of explanation are needed:

The machine was to be housed on two floors; all the equipment that only the maintenance engineers needed to touch - central processor, core memory, controllers etc - would be in the Computer Equipment Room on the ground floor, the input/output units, magnetic tape drives and whatever else the operators needed to use would be in the Computer Room on the first floor, with closed circuit television communication between the two.

Rooms 1-8 on the ground floor were for visiting users who wanted to spend several days in the building, developing a programme and/or running a large computing project. Each had a desk, one chair (deliberately one), a telephone, a filing cabinet and some book shelves, and each could be booked a week at a time, up to two weeks ahead. They were always known as the Cells.

The Think Room on the first floor was a sort of users working common room, where any user could spend up to a few hours in the building.

It proved an excellent building; it met all the needs we had foreseen and was a very pleasant place in which to work. Needs have increased greatly since then, there have been immense changes in the amount and type of equipment installed, and two further wings have been added; but the core of the building remains the same and has accommodated these changes without difficulty. Of course, it was fortunate that it was planned at the time when mainframe computers were such bulky objects that consumed so much power and required such a high-quality environment.

Let me make one rather personal point. As I said, I had a budget for the building, and this included money for furniture and decoration. I have never seen why the working environment should not be at least reasonably civilised. I sought official advise on how I might spend this money and was told that so long as I used it for legitimate purposes and kept within my budget I had a fairly free hand. So instead of the very worthy but decidedly dull standard Government desks and so on I bought modern purpose-designed office furniture that was not only better designed and more attractive but also cost less; and spent a modest amount of the savings on good reproductions of modern paintings - including quite a number if abstracts - to put on the walls. These went very well, especially a very colourful Kandinsky in the machine room; the operators at first looked at this rather askance, but very soon began to appreciate and enjoy it.

The Staff and Services

In 1963 we put out a sort of Manifesto or Declaration of Intent, a little 6-page booklet describing the future laboratory and the services it would provide. I still have a copy. This states categorically that "... it is intended as a national asset and will be available to research workers in all the universities without charge. There is no intention that its use shall be restricted to nuclear physics, even to physical science in general: physicists, psychologists and linguist will be equally welcome in the Laboratory". Later it states as follows:

Like any other computing laboratory, the Atlas Computer Laboratory will have a good deal of service work to do. Many people, from the Rutherford Laboratory, the Atomic Energy Research Establishment and so on, will write their own programs and will only want machine time and the services of operators and data-preparation staff: a large Operations Group is planned to ensure that this need is met - actually, the running of so large and complex an installation poses quite difficult organisational problems. But many prospective users will want help in formulating their problems, in carrying through mathematical developments, in finding or adapting numerical processes and in writing their programs, and the Laboratory will need mathematical staff at a wide range of levels to provide this help: it seems likely that there will be a need for quite a lot of formal teaching of programming, of numerical techniques and of pure mathematics.

It is intended that the Atlas Computer Laboratory shall play its part in exploring (and exploiting) the possible use of computers, and with this end in view will offer research appointments to mathematicians, scientists and others who would like to work in the context of the machine....

Figure 3: The Atlas Computer Laboratory building.

Figure 3: The Atlas Computer Laboratory building.
Large View

This does seem a little ponderous now, but again I beg the reader to remember that it all happened 30 years ago; I hope it will indicate at least that out hearts were in the right place.

I can't resist quoting the following, as a reminder of what inflation has done since then:

Appointments open to graduates and to post-graduates will be in the Scientific Officer and Experimental Officer classes. A first or good second class university degree is normally required for entry to the Scientific Officer class, whose salary scale for the three main grades cover a span from £825 in the basic grade (£1035 after two years' approved experience) to £2695 at the maximum of the Principal Scientific Officer grade.

The laboratory started off with a complement of about 50 staff, organised into three scientific groups, Operations (including Data Preparation) Programming (which would now be called Software) and User Support - the names are self-explanatory - and Administration. Maintenance as I have said, was contracted to ICT. The reference to "research appointments" in the quotation above needs some explanation. I was quite clear that our job was to provide a service, a very sophisticated service certainly but a service nevertheless, and I was very conscious of the risk run by all service organisations, of becoming too inward-looking and forgetting that the service is for the benefit of the users - the classic illustration is the librarian who is happy only when all the books are safely on the shelves. I felt that the way to combat this was to have some able people in the laboratory working on research interest of their own without any responsibility for service work - a sort of leaven - and I was able to get authority for half -a-dozen posts for this specific purpose. These were to be strictly fixed-term appointments, from 1 to 5 years as wished by the holders, and not to be converted into full-time staff posts. We interested several of the Oxford and Cambridge colleges in these posts, and several were set up as joint fellowships with the laboratory and a college - the holders were collectively called the Fellows; in the colleges they were the Atlas Fellows. They did indeed provide the leavening, and proved a great success.

About half the members of the initial staff had transferred from the Harwell computing group, for which I had been responsible. It was my great good fortune that James Hailstone, who had come to Harwell originally to set up and run the punched-card machinery section, came over to head the Operations Group; and that Dr. Robert Churchhouse (now Professor Churchhouse CBE at the University of Wales, Cardiff) joined as the head of the Programming Group - in addition he had strong research interests of his own, in number theory, and made some very interesting applications of Atlas in this field. The "we" in the above almost always means this trio consisting of those two senior colleagues and myself.

Let me say a word about Administration. It's not uncommon for scientists to be rather dismissive about Admin, but in fact this can make all the difference to the smooth running of a scientific organisation. We needed only a fairly small administration group, and we were very fortunate in the people who staffed it, all of whom felt thoroughly caught up in the idea and aims of the Laboratory and worked enthusiastically to help realise these. We were especially fortunate in our excellent secretaries and in our Administration Officer, C.L. Roberts MBE, always known as Robbie. He ensured that the highest standards were maintained and that things got done. A true story I've often told is that one of our contractors once said to me that he'd never known anything so painless - if you were going to need the 25-ton crane outside a certain door at 10.30 am on Thursday week you just told Robbie and forgot about it - you knew it would be there.

Startup

The machine was delivered to the laboratory in April 1964 - in 19 truck loads. Physical installation took a month (there was a lot of plumbing and a vast amount of cabling), when commissioning started and took another 3 months. This required bringing the reliability of both the hardware and the software (operating system) up to the level necessary for an effective service, and I'll not conceal the fact that it gave us plenty of anguish. I well remember our delight when for the first time the complete system ran for two whole hours without a failure - signalling, if I may use a famous statement, the end of the beginning. We felt able to open our doors for regular service in October - five months after delivery. This now sounds terrible, but it wasn't at all bad for the time; I don't think I'm being unfair to anyone when I say that getting every big operating system on the air has proved a harrowing experience of blood and toil, tears and sweat.

I have already said that the long period, two years, between being given authority to set up the laboratory and opening the service gave us time to thinking about the objectives and to plan the operation. It gave us time also to go round the universities and tell them of the services we were planning to offer them and to encourage them to begin to think about how they could use these to further their research work. Bob Churchhouse, Jim Hailstone and I spent a great deal of our time on the road, as it were, and it was time well spent. Not surprisingly, many of the academics looked rather askance at the whole enterprise, suspecting that it was simply draining off funds that would otherwise gone to them; but once convinced that this was something extra and new, set up for their benefit, they became enthusiastic, not simply potential and then actual users of the laboratory, but also collaborators with whom we enjoyed excellent and stimulating relations.

Take-up of the service was fast. Starting with 1-shift working in October 1964, we had to open a second shift two months later and in early 1965 we were running the machine around the clock, 24 hours a day. This continued until we closed it down in March 1973, after 9 years of intensive work. For the record, I should note that in 1965 the Science Research Council (later to become the Science and Engineering Research Council) was formed and took over the ownership and management of the laboratory from NIRNS, which ceased to exist; and in 1968 ICT merged with the English Electronic group of computer companies - English Electric, Leo and Marconi - to form ICL, which thus became the only British manufacturer of mainframe computers.

Early Activities

The purpose of the laboratory was and still is to provide to research workers in academic and certain government institutions computing services on a scale that they could not get from their own institutions. No charge was made for academic work, government users paid a charge that represented the true cost of the work done.

For quite a long time Atlas was much the most powerful computer available for civil work in Britain and quite possibly equal to anything similarly available anywhere else in the world: the only rival was the IBM 7030 ("Stretch"), one of which was installed in the Aldermaston weapons laboratory in 1964. So it is not surprising that demand rose so quickly; I don't have detailed figures for the very early days, but writing for the SRC house journal Quest in 1968 Dr Churchhouse said -

... in a typical week we would run 2500 jobs, input a million cards and 24 miles of paper tape, print two million lines of output, punch 60,000 cards, handle 1500 reels of magnetic tape. We have nearly 1500 projects on our books from university users and are usually doing work on 600 of these.

The analysis of the languages he gives is interesting:

Compiler % of jobs
Fortran 50.6
Machine Code 12.9
Algol 10.6
Extended Mercury Autocode 7.3
Atlas Autocode 3.6
Other 15.0

By that time we had added powerful equipment off-line for graphical output, a Benson-Lehner electromechanical plotter and a Stromberg-Carlson 4020 microfilm plotter. The latter provided an immensely valuable tool; with it we could produce high quality diagrams and pictures, for example of crystal structures, and cine film, for example to show the vibration of an engineering structure; and could generate text in non-latin alphabets, including Greek, Cyrillic, Hebrew and Arabic. For several years it was the only such equipment available to research workers in British universities.

The staff had then risen to about 100. Do not forget - all this work was still "local batch" and came to us on cards, paper tape or magnetic tape, or as manuscript for us to punch.

One or two points are worth making in connection with the list of languages. The dominance of Fortran is evident, reflecting the dominance of scientific application; the compiler was written by a group consisting of Barbara Stokoe and E.B.Fossey of the laboratory and A.R.Curtis and I.C. Pyle of AERE Harwell. It was written in Fortran itself - this must be one of the earliest exploitations of this technique. The Algol was Algol 60, of which there were several dialects being used in the universities. F.R.A. Hopgood (now Chair at Brunel) wrote a comprehensive system that would accept any of these and enabled a lot of people to transfer their work painlessly to Atlas. The experience of a group at Reading provided a striking demonstration of the power of Atlas. Reading, like several British universities at the time, had an Elliott 803 machine for which Algol was the preferred language; this group was pursuing a project for which each run of the programme took about 8 hours on the 803 - a whole shift - and not surprisingly were finding it difficult to get the time they wanted. On Atlas this took 2 minutes, and so became a trivial job of which they could have as many runs as they wanted.

Some Major Applications

With so large a body of users, Atlas supported a range of scientific and other projects that is far too wide to be listed here; but we soon found that there were several major fields of application, in each of which most users' needs could be met by general programs, which we then provided - either writing them ourselves or acquiring (and adapting) them from other sources, usually academic. Among the more important of these were the following.

Crystallography.
Determination of crystal structures by interpretation of x-ray diffraction patterns. Later we added equipment that scanned the films and produced the digital information needed as input to the analysis suite.
Finite-element analysis
Mostly for engineering structures.
Time-series analysis
Used to detect patterns in variations in recorded phenomena such as temperatures, river flows, power flows in the national electricity grid system, elecro-encaphalograph records.
Text analysis
Investigation of quantitative features of natural-language texts, such as word count, vocabulary, frequency distribution of word use, word length and sentence length.
Survey analysis
Operations such as counting, classifying, correlating the information items gathered in a survey undertaken in the study of some medical or social question, such as the incidence or spread of some disease in a particular industry or geographical region.

I should mention also two national projects, space research and weather forecasting. All the processing of the raw data gathered and transmitted to earth by the first three British scientific satellites was done on Atlas; and the Meteorological Office - who had been a strong supporter of the case for establishing the laboratory - developed, over about a year, their new model of the atmosphere, much more realistic and correspondingly more demanding of computer power than anything used up to that time.

Later Years - and Some Reflections

As I said we ran Atlas for nearly 9 years, switching it off for the last time in 1973: that was a long life, during which it had given excellent service and amply lived up to expectations. We had then added a front-end processor (a PDP 15) to provide an on-line terminal system, and a second mainframe machine, an ICL 1906A. We had intended to install also their planned top-of -the range 1908A, but ICL cancelled the project and concentrated on their "New Range", the 2900 series.

When I retired in the summer of 1975 the Atlas Laboratory became part of Rutherford Laboratory, into which the former Radio and Space Research Station (the Appleton Laboratory) had been incorporated and whose name had been changed to the Rutherford Appleton Laboratory. Retaining the Classic name, ACL is now the Atlas Centre of the RAL. The Rutherford's IBM 360/95, bought originally for data processing for high-energy physics, was moved in during 1976 and in turn replaced successively by a Fujitsu Facom 400 (for a time, marketed by ICL under the name of Atlas 10), a Cray XMP-48 accompanied by an IBM 6-processor 3090, and a Cray YMP, the last most powerful mainframe machines available now and at least 1000 times the power of Atlas. Under its present Director, Dr. Brian Davies, it continues to provide top-level computing support to research workers in the universities and other Research Councils; but of course the world has changed, and almost everything now comes into the laboratory (and results go out) over telecommunication links. The laboratory is now so intimately involved with data communication and networking that it houses the headquarters staff of the JANET academic network.

Reflections? - Well, it was a great experience. The Laboratory was and still is a service organisation; this can sound boring, but in fact can be most stimulating. We were providing the very sophisticated services needed by a large body of bright people with a great range of intrinsically very interesting problems; we all learned a great deal by talking with our users - most certainly I did. And the Laboratory was just about the ideal size: large enough, at around 100 staff, to have plenty of internal variety but small and compact enough for everyone to know what was going on and to feel personally involved. Local batch processing did at least have the stimulating feature that there was always great activity in the machine room. I used to like to wander in just to savour this. One day I was watching with interest a visiting user - I didn't know who he was, and he didn't know who I was - inputting some complicated graphical shapes by tracing round the outlines on the D-Mac input table. When he'd finished I asked him what it was all about; he told me, and then asked "Do you work here?" I felt highly flattered.