Contact us Heritage collections Image license terms
HOME ACL Associates Technology Literature Applications Society Software revisited
Further reading □ Overview08/10/6524/04/6630/09/6604/01/6717/07/6717/11/6726/04/6823/07/6804/02/6915/07/6910/12/6903/06/7024/08/7011/11/7031/03/7116/09/7113/10/71 SBSTAR-100 Apr 72
ACD C&A INF CCD CISD Archives Contact us Heritage archives Image license terms

Search

   
ACLLiteratureCommittee MinutesSRC ACC :: SRC ACC Minutes
ACLLiteratureCommittee MinutesSRC ACC :: SRC ACC Minutes
ACL ACD C&A INF CCD CISD Archives
Further reading

Overview
08/10/65
24/04/66
30/09/66
04/01/67
17/07/67
17/11/67
26/04/68
23/07/68
04/02/69
15/07/69
10/12/69
03/06/70
24/08/70
11/11/70
31/03/71
16/09/71
13/10/71 SB
STAR-100 Apr 72

SB 71/11: Future Policy for the Atlas Laboratory

J Howlett

13 October 1971

INTRODUCTION

The question of the future policy and mode of operation of the Atlas Laboratory has been under discussion during the past eighteen months. A first paper, SB 70/24, was submitted to the Science Board in November 1970. The present paper represents the outcome of subsequent discussions, the most recent having been that in the Atlas Committee on 16 September 1971. The Board is invited to consider the proposals and to submit them, modified as necessary, to the Council.

SUMMARY

The Atlas Laboratory has been providing a computing service to university research workers generally since 1964, based on its Atlas computer. During the last few years the activities of the Computer Board have brought about a great improvement in the computational resources of the individual universities, which has changed the nature of the support which research workers most need from the Laboratory. The paper proposes a change to a new mode of operation, in which the Laboratory would give priority to large-scale or otherwise special projects the scientific value of which has been assessed by the appropriate subject committees of the SRC, and for which the cost had been assessed by the Laboratory. The paper advocates also a continuation and expansion of the Laboratory's own programme of research and development. Examples are given of large-scale computing projects already known to the Laboratory and of suitable subjects for the internal programme. The Laboratory is about to acquire considerably more computing power with its recently installed ICL 1906A and its share of the Rutherford Laboratory IBM 360/195, totalling about 6-7 times the equivalent of Atlas, and is considering the need for a new machine of much greater power, the capital cost of which could range from about £2½M to £4M. The change in policy would not significantly affect the predictions of need for staff, accommodation and operating costs already shown in the Forward Look. The paper suggests that the constitution of the Atlas Committee should be reconsidered in the light of the new policy.

I. THE PRESENT SITUATION

BACKGROUND

The Atlas Computer Laboratory was formally set up in late 1961 under the auspices of the National Institute for Research in Nuclear Science and started regular operation in October 1964 with a large Atlas computer system as its main equipment. At that time the British universities were very seriously under-provided with computing power and the idea behind the Laboratory was that it was to be the place to which university research workers could go, when their work needed greater power than their local computer unit could supply. Atlas was then a much more powerful and sophisticated machine than anything else to which university people in general had access. The Laboratory's building, the equipment, the staff numbers and distribution and the entire organisation were oriented towards providing a first-class professional computing service to a large and changing body of users whose interests covered all fields of science and who were physically distributed all over the country. There is no doubt whatever that the original idea was sound and that the services provided by the Laboratory enabled a great number of research projects to be completed which otherwise could not have been started. One incidental outcome has been that the Laboratory effectively set the standards for computer accommodation, staffing and operating in British universities.

EQUIPMENT

The Laboratory's main equipment now is as follows:

  1. The original Atlas computer, with some enhancements; capital cost £2.8M.
  2. An ICL 1906A system, of which the raw computing speed is about three times that of Atlas. This was delivered in July 1971 and is now being commissioned; capital cost £1.4M.
  3. An SDS Sigma-2 computer, linked to Atlas through the latter's disc store, and used at present to control the multi-access system developed by the Laboratory; this has 28 independent terminals, mostly located within the Laboratory. The total capital cost of this system with all the terminals was £l67K.
  4. A computer graphics system comprising a PDP-15 computer controlling a VT-15 programmable display, a Stromberg-Digigraphix Type 4020 microfilm recorder and a D-MAC graphical input table. The SD 4020 has been in use for three years, and has been the only equipment available to university people for producing computer output, either text or graphical, directly on microfilm or hardcopy. The whole complex will be linked to the 1906A in 1972. The total cost here was £170K.

Around the end of 1971 we shall install an IBM 1130 small computer (cost £85K) to act as input/output station to the large IBM 360/195 which is to be installed in the Rutherford Laboratory during December. The Council has ruled that we shall have 20% of the time of this machine to allocate to users outside the nuclear physics field, and this share should be the equivalent of about four times the computing power of Atlas.

We are planning to close down Atlas in late 1971 and are discussing possible future machines: I return to this point in Section VII.

THE SERVICE TO UNIVERSITIES

The computing service has of course been provided solely by Atlas and will continue so until the 1906A and the 360/195 are in operation. About 70% of the machine 's time has gone to university research workers and the rest has been divided fairly evenly between our own internal work and work for government organisations, including other Research Councils, who pay for all the work done for them. The annual receipts from this source have been about £100,000.

The university work clearly dominates. The method by which it comes to us is this. We have a local representative in every university, who is either the head of the university's own computer unit or his nominee; he gets all our literature and is the official link with the Laboratory. Anyone wanting to get work done on our machine must fill in a job application form describing the job and giving the amounts of time and storage needed and other details, and submit this to his local Atlas representative. The latter can raise any questions or objections he wishes; for example, if his own machine could meet the needs he should say so. If he is prepared to support the application he countersigns the form and sends it to us. We may query the application - we may feel, for example, that the demand was excessive in some way, or that the work could be done with a standard program - but if we are satisfied we issue a job number which allows the man access to our system and from then on he deals directly with us. However, we send a weekly accounting statement to every representative showing the amount of time used and the notional cost of the work done under every job number for which he has signed and the totals to date, so that he is kept informed of the way in which his university is using our resources.

This system is simple and direct and has worked very well. It has, however, three features the effect of which is to make the Laboratory's operation very similar to that of a university central computer unit; these are that, in general, (there are exceptions)

  1. we do not select projects;
  2. we give no priorities nor guarantees;
  3. we rely on the university representatives to assess the soundness of the applications.

As I said, there are exceptions; we will give priority to someone who really has to meet a deadline and we have sought independent opinions on the scientific quality of some big demands and have criticised programming methods. But in the main this is the situation, and the aim of this paper is to show that a change from this way of working would be more appropriate in today's circumstances.

An indication of how the present policy has worked out is given by the following table. This refers to the period 13 September 1963 to 30 July 1971 (roughly the past 2½ years) and the subject headings are those given by the applicants in classifying their work:

Table 1

Total Number
of Users
Total Time
Hours
Number of
Users taking
>50 hours
Time
hours
Mathematics 112 2,618 8 1,764
Physics 150 3,351 14 2,153
Chemistry 222 5,628 33 4,099
Engineering 207 3,257 21 1,593
Chemical Engineering 10 321 2 243
Medicine 39 402 1 58
Meteorology 3 208 1 207
Other 251 1,367 3 207
Total 994 17,152 83 10,324

The concentrations into Physics and Chemistry is only to be expected; 'Mathematics' certainly includes a good deal of work which could be described as theoretical physics, or even engineering, but one would have to look through all the individual applications to get the details. It is also to be expected that long-running jobs take up a greater fraction of the time than is the case for a university computer, but it is perhaps a little surprising that the 83 projects taking over 50 hours each (8% of the total) should account for 60% of the total time.

STAFF AND ORGANISATION

The complement for 1971/72 and the bid for 1972/73 are as follows:

Table 2

Group 1971/72 1972/73
Director's Office 2 3
Operations Group 69 72
Programming Group 27 31
Support Group 14 17
Research Posts 5 6
Administration 19 19
Total 136 148

The Operations Group is responsible for everything involved in actually running the service - operating the computers (three shifts, with extensions into the weekend), card and tape punching, reception and dispatch, control of the magnetic tape library and so on. With a total of about 1,000 users, all away from the site, and about 3,000 jobs run each week, this is a large-scale operation. That it has gone so well is due in quite a large measure to the fact that enough complement places were provided from the beginning to enable all the necessary activities to be properly manned.

The Programming Group is staffed almost entirely by graduates, mostly in mathematics or mathematical subjects. There is a need for a continuing software effort in support of any computing service, especially in one as large and as varied as ours, and in our environment of advanced scientific research it is essential to have a professional staff of very high quality in order to keep in touch with, and to exploit, the latest developments in computer science - using that term in its most general sense.. This group has produced much of the basic and applications software used in the Laboratory, about which I shall say more in the next section.

The Support Group provides as it were a bridge between the user and the Operations Group. As a general principle we do not write programs for individual users, but we will help users to understand and get the most out of the machine's operating system, the many operational facilities which are available and the library of programs and program packages which we maintain. We will also help them in the development of their programs by monitoring test runs, and will oversee long sequences of production runs. This service is greatly appreciated by users and also contributes a good deal to the efficient use of the computer. It is exacting work which demands a great deal of skill and knowledge of programming and also a continued willingness to interest oneself in the details of other people's work.

The Research Posts were established to enable us to have people of high ability working in the Laboratory on their individual research projects, who needed the kinds of resource which the Laboratory provided. These appointments are always made for a fixed term usual1y three years. We have had the good fortune to get the cooperation of several of the Oxford and Cambridge Colleges and have been able to associate some of the posts with Research Fellowships. The Colleges involved have been Pembroke, St Catherine's, St Hilda's and Trinity at Oxford, and Churchill at Cambridge. Amongst our appointees have been four pure mathematicians (two number theorists, two algebraists), three applied mathematicians (one working on radiative transport, one on atmospheric physics and planetary atmospheres and one on integer-programming and related techniques used in operational research), a numerical analyst, an engineer and an archaeologist - the last interested in the application of the computer to classification problems. I believe that this has been a very valuable feature which both enlivened the Laboratory and helped to strengthen our links with the universities.

Finally, the Administration Group provides general services which are needed by the Laboratory as a whole. We use the Rutherford Laboratory wherever possible, for example for all personnel work, accounting and bill paying, building and plant maintenance (not the computer, of course); our own group provides those services which must be located within the Laboratory, such as general clerical and typing services, local supplies and finance, stores and cleaning.

INTERNAL WORK

This refers mainly to the Programming Group and divides broadly into basic and applications programming.

By basic software I mean the corpus of programs which, so to say, make the whole system work: control the flow of work through the machine, enable it to accept users' programs written in high-level languages, provide the means for communication or interaction between the user and the machine, provide output in the form required and provide information about the running of a program which will help in the detection and correction of errors and in the improvement of its efficiency. A vast amount of such software is needed to support a varied service on a big modern computer and one has to rely on the manufacturers for much of the greater part of this, certainly for the operating system and for the compilers for the main languages such as Fortran and Algol. It is essential in a large laboratory such as ours to have basic software experts who understand the workings of the manufacturer's programs (otherwise one can be operating a very expensive installation blind) and who can design and produce whatever is needed in addition to what the manufacturer provides. The main products of the Laboratory staff have been:

  1. the Fortran system for Atlas - a joint undertaking with AERE Harwell, with assistance from the original maker of Atlas, Ferranti Limited;
  2. an exceedingly flexible and comprehensive Algol system for Atlas, based on the ICL compiler;
  3. a number of compilers for special languages including LISP (list processing), SOL (simulation), SNOBOL (string and character processing);
  4. the utility routines for Atlas, which give the user a wide variety of data-handling facilities such as storage on private magnetic tapes, editing, selective printing, sorting and merging;
  5. the multi-access system;
  6. the whole of the software for graphical input (D-MAC) and output (SD 4020) equipment and for the PDP-15/VT-15 interactive graphics system, including programs for the production of output as cine film.

All of these were considerable undertakings, and (i), (ii), (v) and (vi) - the last still in progress - have been major projects.

Applications software, as the name implies, is aimed at solving particular problems or classes of problem. We have had a positive policy of identifying areas of science in which large, general purpose programs or packages (connected sets of programs) would meet the needs of at least the majority of workers, and then producing the programs either by writing them ourselves, by acquiring them from other sources, by developing them from existing programs or by getting them written under contracts. These are always very big programming projects, often continuing in stages over several years, and in fact usually have to be on this scale to give the flexibility and wide coverage which are needed; but the effort is repaid by their long-term value to very many users and the saving of individual programming effort. The principal systems of this kind which we have made available deal with:

  1. Crystallography
  2. Quantum Chemistry
  3. NMR Spectroscopy
  4. Time-series Analysis
  5. Statistics Survey Analysis
  6. Information Storage and Retrieval
  7. Text Analysis

(These are of course in addition to the large library of standard mathematical routines, such as matrix processing and numerical integration.) We provide very comprehensive supporting services with these systems; in crystallography, for example, the Laboratory has become effectively a national computing centre.

A major effort of a rather different nature has been the processing of the data from space satellites. We collaborated in the programming in the case of UK- 3 (renamed Ariel 3) and are wholly responsible for this in the case of UK-4 and of the S-68 experiment; and we have just agreed to take on the processing of the GALAXY output for the Southern Sky Survey. The greater part of this work has in fact been done by the Support Group. This is referred to again, with more detail, in Section V.

SUMMARY OF THIS SECTION

My main aim in giving all this information about the Laboratory's activities is to show that it is a very powerful organisation with great resources of expertise in all fields of computer science and computer operation. A most valuable hidden resource is the quite unrivalled set of contacts with the university world. Computers are powerful devices of apparently unlimited generality, and in general the people who work with than are almost equally adaptable. Thus the Council has here a powerful weapon the use of which is entirely under its control and which can play a major role in the Council's policies for the support of science.

II. THE CHANGED CIRCUMSTANCES

The Joint Working Group on Computers for Research - better known as the Flowers Committee - was set up by the Secretary of State for Education and Science in 1964 with the terms of reference To assess the probable computer needs during the next five years of users in universities and civil research establishments receiving support from government funds. The report (Cmnd. 2883, published January 1966) made plain the serious lack of computing power in British universities, proposed a coherent plan for equipping them adequately and led to a government agreement to finance a slightly modified programme and to the formation of the Computer Board to implement this programme. To date the Board has spent or committed a total of about £30M on capital equipment for university computer centres and has contributed about £6M to running costs. This has completely transformed the situation. Most of the universities now have at least reasonable computing equipment, properly housed and staffed, some are very well equipped and the few which are still underpowered will have their position improved quite shortly. The situation is of course not static; demand increases all the time and the Board has a programme of continual enhancement of its installations. At least three centres - London, Manchester and Cambridge - have or will soon have much more power than the Atlas Laboratory.

The Laboratory is thus in a very different relation to the universities from what it was in 1964. Much more computing power and expertise are now available within the universities and they can satisfy much more of the general demand with their own resources. They do still find it difficult, often effectively impossible, to meet the needs of research projects which make large-scale demands on time or storage or require special hardware or software facilities. The main thesis of this paper is that whilst the need for general support outside the universities has declined, the need for selective or specialised support has increased and that the Atlas Laboratory, working in close collaboration with the Council's Boards and Committees, is uniquely well suited to provide this.

III. DEMANDS FOR LARGE-SCALE COMPUTATION

It is very difficult indeed to predict with anything approaching precision the future needs for computing in support of research. This is partly because the course of scientific research itself is not predictable in any long-term sense, partly because every new advance in computer power and facilities has opened up new and exciting possibilities of attack. Professor Burke wrote this in a recent letter to Mr Jolliffe: It is generally accepted that current understanding of Nature has reached the point in many fields where the systems which are studied are so complex that current analytical theories and laboratory experiments are inadequate. For example, electron scattering by highly ionised atomic species which are basic in our interpretation of astrophysical observations can only be studied on a computer. Another example is the solution of many-body non-linear problems with complex boundary conditions such as occur in nuclear fusion devices. In both these areas very important computational developments have occurred in the last few years which open the way for extremely rapid progress in the future. There is also the whole field of 3-dimensional physics waiting to be studied, for although we live in a 3-dimensional world we are still virtually unable to tackle realistic 3-dimensional problems. A rather extreme example of a proposed attack is the intention of NASA to use the somewhat fabulous ILLIAC-4 machine to compute the aerodynamics of an actual aeroplane; if successful this could lead to the replacement of many wind-tunnel measurements by computation, almost certainly at much lower cost.

It seems that the main fields in which good and important science is to be done, needing computation on a large scale, are:

A Working Party on Computational Physics, under the Chairmanship of Dr K V Roberts of the Culham Laboratory, met during 1969/70 to try to assess demands. A summary of the findings of this body was given as Appendix B to my previous paper (SB 70/24) on the Laboratory's policy. No report was issued, but the informal estimate made at the time was that there would be a need beyond what the universities could provide with their own machines equivalent to 10-20 times the computing power of Atlas. The need for very large amounts of storage, powerful and sophisticated peripheral devices (eg graphical output) and effective means for communication and interaction with the machine were stressed.

Two Working Parties under the chairmanship of Dr Francis are studying the internal needs of the SRC and of the other Research Councils, and will also attempt to get an estimate of the part of the university need which SRC might be asked to meet. These groups will not be reporting until early 1972, but it is possible that provisional information will be available for presentation at the Council meeting in December.

We ourselves are in process of consulting members of the Boards and Committees who are themselves concerned with work needing large amounts of computation. '!here will certainly be some authoritative views available in time for the meeting.

IV. FUTURE POLICY FOR THE LABORATORY

I should like the Board to consider the proposal that the computing service which the Laboratory provides to universities should change gradually from the present form to a more selective one oriented more towards projects needing computing on a large scale or other special facilities. The procedure would be as follows.

  1. A research worker with such needs would apply fairly informally to the Laboratory for a first discussion of his project and the possibility of the Laboratory meeting his needs. Depending on the scale and complexity of the project there would be further discussions with the members of the Laboratory best able to advise him and possibly with one or more members of the relevant SRC subject committee, leading to a precise statement of the computational problem and of the demands to be made on the Laboratory. This kind of discussion between prospective users and members of the Laboratory is already a common feature of our work.
  2. The man would make a formal application to the SRC, just as he would for any other kind of grant. A copy would go to the Laboratory who would give an estimate of the cost of the work requested, using an agreed costing procedure, and would say whether or not the amount of time requested, and the rate of take-up, could be provided.
  3. The application would be considered by the appropriate subject committee and judged on its scientific merits; as with all other applications, the committee could award the full amount asked for, or some reduced amount or reject. The committee would take into account the Laboratory's statement of cost and of availability in making its decision.
  4. The Laboratory would guarantee to the applicant the service recommended by the Committee and from then on the applicant would deal directly with the Laboratory. The usual rules about reporting progress to the Committee would apply.

Any time not committed to projects which had been approved in this way, or to the Laboratory's own internal work or to paid work for other government organisations, could be used for smaller-scale work applied for directly from universities as at present. Anything accepted in this way would carry no guarantees, although the Laboratory would of course give the best service it could.

There are good reasons for leaving some fraction of time, say up to 20%, always uncommitted, so that the Laboratory could always give some service to people with more tentative or informal needs, or who wanted to get something, necessarily on a relatively moderate scale, done more quickly than the committee procedure normally allows.

This method of procedure would have a number of advantages. It would enable a research worker to plan a large-scale, long term project with the knowledge that if its scientific merit was accepted he would be assured of the computing facilities he needed in order to carry it through to the end. It would ensure that the work done in this way was of the scientific quality which the Council was prepared to support. It would enable the Council, through its Boards and Committees, to use the Laboratory in a positive way to support significant lines of research - and to take chances on far-out projects if it felt this worth the risk. And it would exploit the special position for the Laboratory which, whilst it is deeply involved in the academic world and committed to serving the needs of research, is not tied to any individual university or any particular project and can therefore choose how it shall use its resources.

The change to such a policy would necessarily be gradual and evolutionary. The universities as a whole rely on the present Atlas service to supplement their resources and many users have big investments in programs for this machine. I should like to see this service continued until Atlas is closed down, probably near the end of 1972. The natural time to start introducing it is early 1972 when the 1906A and the 360/195 come into service, for there is no reason why we should use these machines in the same way as we have used Atlas. The development of the new style of service would have to be kept under review and changes made as experience suggested. The Atlas Committee could give rulings on such parameters as the fractions of time which should be committed to approved university projects, internal work and external (eg government) users. I have referred to the Laboratory's internal work in Section I and consider it further in Section VI. This is a convenient place at which to say that I should expect to make an annual submission to the Science Board, through the Atlas Committee, stating the main projects on which we proposed to work and their needs in manpower, equipment and computing services.

VIEWS OF THE SRC BOARDS AND COMMITTEES ON THE PROPOSED POLICY

The previous Science Board paper on this subject (SB 70/2A) was sent to the other Boards and to the subject committees for comment. The comments received can be summarised as follows:

  1. The Science Board itself gave provisional approval. The committees for Biological Sciences, Chemistry, Enzyme Chemistry and Technology and Mathematics welcomes the proposals and expressed no reservations. The Neutron Beam Research Committee agreed that London Office should organise a discussion with Dr Howlett to investigate the possibility of wider use of Atlas facilities by neutron beam users. This has been done, and agreement reached on the use of the 360/195. The Physics Committee was in favour of the proposals.

  2. The Astronomy, Space and Radio Board, and relevant committees, reported as follows:

    1. The proposals were generally welcomed.
    2. The right balance would have to be achieved between the facilities available - and the use of them - at the Atlas Laboratory and universities or other SRC establishments. Intermediate jobs should not fall down between the two.
    3. Fairly large demands could be expected from the astronomers and space scientists.
    4. The need for data links should be considered further.
    5. The proposals might create the need, initially at any rate, for more software, ie reprogramming from existing computers. This need could be considerable (though it was not in any way advanced as an argument against the proposals).

    The comment was also made in connection with (ii) that a distribution of computer power avoids the worst consequences of a serious mishap at a very large computer centre.

    The Astronomy, Space and Radio Board hoped that in developing his proposals Dr Howlett would talk to the Royal Greenwich Observatory, the Royal Observatory, Edinburgh, and the Radio and Space Research Station as well as to the large university groups that the Board supported. There has been much consultation with the Astronomy, Space and Radio establishments, leading to the projects described in Section V(6).

  3. The Nuclear Physics Board expressed reservations about the proposals. An appreciable amount of nuclear structure computation was handled through the Atlas Laboratory, and the Board were anxious that small users should not suffer as a result of any change in policy.

  4. The Computing Science Committee of the Engineering Board had given some consideration to the proposals and while the committee had no collective view it considered that Atlas, although no longer occupying the unique position it enjoyed in earlier years, would continue to fulfill a real need in universities. The growth of teaching in computing in universities would place an additional burden on universities' computing facilities and this could have considerable implications for the future nature of the services provided by Atlas. Further specific proposals by the Computing Science Committee are described in Section VI.

    Although the Engineering Board subsequently endorsed the proposals, it added that there appeared to be only limited scope for the use of the Atlas Laboratory for engineering grants.

IMPLICATIONS FOR THE ATLAS COMPUTER COMMITTEE

The Atlas Committee, under its successive chairmen Lord Penney, Sir Brian Flowers, Lord Halsbury and Sir Rudolf Peierls, has guided the laboratory's operation and development over the past ten years. I am most grateful for all the help and support which they have given me so freely throughout this time. However I feel that if the Laboratory is to become more closely involved with large projects from the whole range of science which the Council supports, there will be a serious need for something of the nature of a scientific advisory body. This is not intended to usurp the functions of the subject committees, who must be the judges of the scientific merits of the projects submitted to the Laboratory, but to keep the Laboratory alive to the developing fields from which worthwhile computational projects should come and if necessary to advise on the resolution of questions of priority.

This aim could be achieved by broadening the membership and the terms of reference of the Atlas Committee. New members could be chosen to advise on particular, if broad, fields of science - for example, theoretical chemistry, solid-state physics, structural engineering; and the terms of reference expanded to read, say, as follows:

  1. to advise the Science Board on the operation of the Atlas Computer Laboratory; and
  2. to advise the Council, in consultation with the Boards, on priorities in the undertaking of work at the Atlas Laboratory requested by the Boards.

I should like to have the Board's advice on this question.

V. EXAMPLES OF LARGE-SCALE PROJECTS

This Section outlines a few specific large-scale problems, to show the kinds of project which research workers are formulating and for which they cannot get the required computational support in their universities. All of these have either been submitted already to the SRC or discussed informally with us in the Laboratory.

  1. Professor P G Burke (Queen's, Belfast). Calculation of scattering cross sections for electron/atom, electron/molecule and atom/atom interactions. The computational problem is the solution of sets of linked integro-differential equations derived from the Schrodinger equation. Much work on these problems has been done on the ICL 1907 machine at Belfast but that is saturated and quite unable to provide the time which is needed; Belfast will not be able to install a new machine before 1973. Professor Burke estimates that his program of work needs 10-20 hours a week of 1906A time in 1971/72, 20-30 hours a week in 1972/73 and 30-40 hours a week in 1973/74. It also needs at least l28K of core store and about 10M characters of private storage on disc. Professor Burke has submitted this to the Physics Committee.
  2. Professor R W Hockney (Reading). A project to extend the particle model work started by Professor Hockney in New York, to study a number of physical systems the properties of which depend on the collective behaviour of interacting particles. Proposed studies include instabilities in hot gas plasmas, evolution of structure in galaxies, the motion of holes and electrons in semiconductor devices and molecular dynamics of phase-changes.

    Professor Hockney needs to use the IBM 360/195 for this work and estimates a need for 3-4 hours a week for five years; he has a strong need for the use of microfilm for output. Professor Hockney has submitted this to the Physics Committee.

  3. Dr K Singer (Royal Holloway College). A large-scale, long term theoretical study of the physics of amorphous materials by computer simulations using particle models. Dr Singer put an application to the Chemistry Committee asking for support for a number of projects under the general heading, which in total required 70-80 hours a week of Atlas time or its equivalent for five years. The Committee was very favourably impressed by the application but decided to support part of the complete project for three years and to consider extending this after the work had progressed some way. We are expecting to provide about five hours a week on the 1906A, Dr Singer getting the balance of his needs from one or other of the London University machines.
  4. Dr B T Sutcliffe (York). To quote: Our aim is to produce very accurate electronic wave functions, within the Born-Oppenheimer approximation, for systems comprising 10-20 electrons, not only for isolated molecules but also simple dimers. We hope to be able to produce such functions not only for ground states, but for one or two low lying excited states and not only at equilibrium values of the nuclear parameters but at sufficient values to obtain an adiabatic energy surface where appropriate. Dr Sutcliffe estimates a need for about one hour a week on the 360/195 for 2-3 years. He has put an application to the Chemistry Committee.
  5. Dr M Petrovic and Dr G Kuo-Petrovic (Oxford). Their interest is in 3-dimensional field problems in general, using finite-difference methods to solve the partial differential equations. As a part of the attack on the problems they am to develop a language and a translation program by means of which the computer program for a particular problem can be generated from a fairly natural statement of the equations. They have had preliminary discussions with us on the possibility of studying mathematical models of the magnetosphere of a pulsar, for which they would expect to need up to two hours a week on the 360/195 for 1-2 years. They are also considering applying their methods to the solution of problems of viscous flow, but have made no specific proposals. Since these problems involve the processing of very large matrices, there is a strong need for large amounts of storage.
  6. Space Research Data Processing. This of course is part of SRC's own scientific programme. The essential feature is that a very large amount of raw data is produced by equipment launched with a space satellite, and that this requires elaborate computer processing in order to derive the scientific information for which the whole project was undertaken. The Laboratory has been or will be heavily involved in four major projects.

    1. UK-3 (Ariel 3): The satellite was launched in mid-1967 and heavy data-processing started in November 1967. All the work has been done on Atlas; the load was 3-4 hours a week at the start, rose to 5-10 hours a week in late 1968 and fell to a very light load in early 1971. The project is virtually completed apart from general tidying up. A very large amount of consequential work has been done for the experimenters on the information with which we provided them, totalling about 1,000 hours of Atlas time and making much use of the microfilm recorder.
    2. UK-4: This is due to be launched in November 1971 and processing should start in December. The load is expected to be about five hours a week on the 1906A and the experiment is scheduled to run for a year; experience with UK-3, which also was scheduled originally for a year, leads one to expect that it will in fact run for a good deal longer.
    3. S-68: This experiment is to be launched in the ESRO TD-1 A satellite in February 1972. The initial separation of the raw data will be done at Darmstadt. The load is expected to be 3-5 hours a week on the 1906A.
    4. Southern Sky Survey. This is not a satellite experiment, but an astronomical survey. The GALAXY machine at the Royal Observatory, Edinburgh, will scan the photographic plates and record the co-ordinates and other information on magnetic tape of every star detected. We have undertaken to do the subsequent processing of this raw information, which is in effect the setting up of a data bank and the production of star catalogues. The survey is due to start in late 1972. The equipment will produce 108 bits a day and the survey will take about two years, so the total amount of information to be stored is between 1010 and lO11 bits, equivalent to about 1,000 reels of magnetic tape. The present estimate of time needed is about twenty hours a week on the 1906A. It is clear that this project is going to need a very careful study of the way in which the information is to be stored, both for the sake of compactness and in the interests of subsequent retrieval.

    All this work calls for close co-operation with other parts of SRC - the Radio and Space Research Laboratory at Slough, the Astrophysics Research Unit at Culham and the observatories. All the projects present problems of file handling and data storage.

VI. PROGRAMME OF INTERNAL WORK

As I showed in Section I, the Laboratory needs to have a vigorous programme of high quality work of its own, both for the direct support of the computing service and to ensure that it keeps up the developments and maintains a high professional standard. The high standard of the service in the past and the high efficiency with which the computer has been operated have been made possible only because the Laboratory has built up strong reserves of skills and expertise. The need will be even greater in the future if the Laboratory becomes increasingly involved with users whose needs are complex and on a large scale.

In a fast developing field it would be only too easy to find that one had become exceedingly good at doing something which no one wanted any longer. Thus in the apparently mundane case of input to the computer, it seems certain that punched cards and paper tape will be replaced by document readers and direct-entry devices. Also we shall not be able to operate our service much longer by relying almost wholly on the mail for receipt and dispatch of material - the turn-around is already too slow; we shall have to make much more use of data links. All these changes have important implications for hardware and software and for the staffing of the Operations Group, and we need to have sufficient knowledge within the Laboratory to prepare for such changes and to take advantage of the new opportunities which they offer.

It seems equally certain that the need for large comprehensive programs and packages such as the crystallographic system, or the graphics system, will increase, especially if we pursue the policy of specialisation and selectivity which is proposed in Section IV. It is a full-time, professional and very skilled task to design, construct, implement and maintain software on this scale, which cannot be undertaken unless staff of the right calibre are available.

The foregoing has referred to work which is clearly and directly in support of the service. However, the SRC recognises computer science in general as a subject which it wishes to support and maintains a special subject committee for the purpose. The Laboratory has not interacted very much with this committee in the past, mainly because of the pressures of the service work, but could do so and should, in my view and that of many of the staff. It would not be appropriate for the Laboratory to concern itself with abstract branches of the subject such as theory of computability, which are much better tackled in universities, but much of it is aimed at very practical ends and could very well be studied in an organisation which has to work under the discipline of a large-scale regular computing service.

The Computing Science Committee has recently been looking into the future needs of the subject and set up two Panels, one under Dr R M Needham of Cambridge for Long Range Research, the other under Professor R A Brooker of Essex for Applied Research. Neither attempted any estimates of the size of the demand. The second, which I feel is the more relevant to the Atlas Laboratory, proposed the following as the most important lines for research and development:

and suggested that the Atlas Laboratory should become involved in the first two.

We ourselves already have interests, with varying degrees of involvement, in the following:

We probably have the best equipment in the country for the last of these and are in a uniquely favourable position to develop computer animation as a powerful aid in education and in science.

I believe that the Laboratory has the potential to become one of the world's leading centres for practical computer science and that a deeper involvement in the subject could only improve the quality of the service which is its first obligation.

To conclude this section, I quote two passages from the Report of the Computing Science Committee (Appendix 2 to CS/Y/13 70/71 May 1971) referred to above:

4.3 Communications

There is ample evidence that communications will continue to grow in importance: too little research effort is being devoted to the relevant problems.

SRC should encourage and partly fund a communications network of the type in which several diverse computers are linked together; the problems in this area are quite different from those of the systems in which many small terminals are connected to one central computer complex. The reasons for wishing to connect computers together in a network include back-up, shedding of peak loads, shared access to special facilities (including special data-bases), transmission of information and the possibility of sharing research projects between distant teams. All of these reasons are potentially important, and inevitably there will be a need for knowledge and experience in creating and operating such networks.

One of the nodes of such a network could be at the Atlas Computer Laboratory, with other nodes at universities and regional centres. The project need not be very expensive if it uses existing machines and if the network is live for only a few hours in each day. The experience gained would be difficult to obtain in any other way: the South-West Universities project, linking computers at Bristol, Exeter, Bath and Swansea, has some of the right characteristics, but not all; in particular, the computers involved are all of a single type.

4.4 Parallelism in Hardware and Software

A generation of large machines is being born in which high speeds are attained by simultaneous operation of many computing elements: these machines are different in concept from conventional serial systems, and demand different concepts in operating systems and in programming. These new ideas will inevitably filter down to common use in smaller, conventional systems of the future, and their understanding will form an indispensable part of the resources of the computing industry.

SRC should support a research project to look at a variety of machines, both large and small, incorporating various forms of parallelism, possibly purchasing one or two of them for a group or groups to conduct individual investigations.

An appropriate centre for such a project might be the Atlas Computer Laboratory. The Laboratory already has proposals for acquiring a large machine (such as a CDC STAR), and has experience of providing a good environment for visiting research workers. Some possible investigations would be: assessing the merits of various machine architectures, detail monitoring of the performance of machines in terms of hardware and software operations, tuning the machine to optimise throughput.

VII. IMPLICATIONS FOR EQUIPMENT, ACCOMMODATION, STAFF AND FINANCE

EQUIPMENT

The dominant equipment of the Laboratory is its main computer. The level of computing power which should be installed is determined by the needs of its legitimate users and as I showed in Section III it is exceedingly difficult to make any precise assessment of this need and of its rate of growth. The examples already quoted show that there are now, ready and waiting, projects of good scientific value, in fields which the SRC is supporting, needing computation on a larger scale than their originators can get from their own universities.

The Atlas Committee discussed this question at its meeting on 16 September 1971. In the paper which formed the basis of the discussion, ACC 71/13, I had given my view that there would be a need to have a machine at least in the 10-30 × Atlas range installed and working by 1974, and that this should be either initially of greater power or capable of enhancement beyond this. The Committee unanimously supported this view, but felt that the real need would be much nearer to the top end of this range than the bottom. Their general view was that the Laboratory should have the most powerful machine which the SRC could afford and that, apart from the necessary raw power, it should have architecture carefully chosen to match as well as possible the type of work to be expected. They noted in particular the CDC STAR, which has special vector-processing operations which might be of particular value.

I had suggested three machines for the Committee's consideration, as follows:

  1. The ICL P4, from its project New Range which is to replace the 1900 and System 4 ranges. About 12-15 × Atlas, cost £2-2½M.
  2. The CDC 7600. About 20-25 × Atlas, cost £3-3½M.
  3. The CDC STAR. Probably 30-50 × Atlas, cost about £4M.

I reproduce in Appendix A the section of ACC 71/13 dealing with these machines, and in Appendix B the relevant Minute of the meeting.

To summarise, I am suggesting that there is a need for a new major computer to be working in the Laboratory in 1974, which could cost between about £2½M and £4M, depending upon the machine and the configuration chosen. We are in process of collecting estimates of need on which to base as precise a statement as possible, and expect to be in a position to put a formal case to the Board in about six months' time.

STAFF, ACCOMMODATION AND FINANCE

The following is based on the relevant part of ACC 71/13, discussed by the Atlas Committee on 16 September 1971. The views were put to the Committee, not as formal bids, but as indications of how I felt the Laboratory should develop under the proposed new policy.

Staff

The numbers and distributions for 1971/72 and 1972/3 are given in Section I, and in the Forward Look we have shown a levelling-off at 150 in 1972/73. At present the Operations Group is much the biggest of the functional groups into which the Laboratory is organised, but we expect this situation to change over the next few years. As I said in Section VI, we shall be making much more use of data links and remote input/output terminals and of alternatives to punched cards and paper tape for primary input. The Head of the Operations Group has said that he expects the job of running a big computer to become more like that of an air traffic controller, done from a desk with elaborate visual displays on which any necessary information can be called up. At the same time there will be a need for a bigger Programming Group, to deal with the much more complex systems with which we shall be involved and allow us to give more support at a higher level to our large-scale users. We are already feeling the need for this and have transferred complement places from the Operations Group. I feel that the estimate of 150 is still valid, and I would suggest the following tentative distribution:

Directorate and Administration 22
Operations 60
Programming and Research posts 50
Support 18
Total 150

I have put the Programming and Research posts together because I can visualise the distinction between the two becoming less marked; the boundary between Programming and Support may also become less well defined and there could be some flow across this.

This is not intended as a bid for this complement but an indication of how I foresee developments. I shall give a much fuller argument when I have to make firm proposals.

Accommodation

With the new computer building and the space released when Atlas is removed we have plenty of accommodation for any machinery I can foresee. Same internal rearrangement might be necessary and possibly some changes to the air-conditioning plant, but I cannot see need for any further major building work on this score.

There will be a need for more office accommodation. If we have a staff of 150 distributed as above, we shall need office accommodation for about 90 - all of the Programming Group (50), 15 from the Directorate and Administration, 15 from the Support Group and 10 from Operations. We should allow also for vacation consultants and other university people attached temporarily but for long enough to need proper office space; I suggest 10, giving a total of 100 people to be housed. The Rutherford Laboratory uses an average of 110 sq ft per person in assessing office space, so the need is for about 11,000 sq ft. The present building provides 6,500 sq ft for offices, so the further requirement is about 4,500 sq ft. We believe that a simplified version of the office building proposed in ACC 70/9 could provide this accommodation at a cost of about £80,000.

Finance

Capital costs depend entirely on decisions about a new computer. The total cost could range from about £2.4M to £4M or possibly rather more; the form and timing of payments would have to be negotiated with the manufacturer, but almost certainly could be spread over three or four years.

The Forward Look has been based on a growth to 150 staff and on costs for computer maintenance and other services which should not change greatly in the proposed new circumstances. Therefore the recurrent cost of running the Laboratory should not be significantly different from what has already been forecast.

The Committee supported these views, agreeing that the estimates were reasonable. In particular, they agreed with views that fewer people would be needed to operate the advanced machines of the future, whilst more would be needed for software development and related work.

APPENDIX A

Extract from ACC71/13: Future Development

POSSIBLE COMPUTERS (paragraph III of the paper)

Assuming that Atlas is closed down in the second half of 1972, we can take it that the Laboratory's computing power is provided by the 1906A and its 20% share of the RHEL 360/195. The latter is likely to be equivalent to about 4 × Atlas, and for some time at least will be our main source of computing power. The 1906A is rated by ICL as about 2½ × Atlas. There is not yet enough operational experience of the machine to provide a firm assessment, but the indications are that this will not be far out; certainly 2 × Atlas should be safe.

If we can take it that the Laboratory needs its own machine with at least computing power in the range 10-30 Atlas, we have to choose between the ICL New Range P4 (referred to in past literature as P52) and one of the big American machines; I should like the Committee to consider two CDC machines, the 7600 (now being marketed as Cyber 76) and STAR-100.

ICL P4

This is not yet announced but ICL have given several confidential presentations to Government bodies (eg the Computer Board) and in May 1971 gave a three-day detailed technical presentation to a group from the Laboratory. The New Range is intended to replace the 1900 and System 4 ranges and will comprise four compatible processors P1-P4, of which P4 is the top, for the present at least. The design draws a great deal from Professor Kilburn's MU-5 project; it is oriented to the use of high-level languages and to the development of multi-processor systems. Professor Kilburn will be able to tell the Committee anything they may wish to know about this design.

ICL say that the raw hardware speed of the P4 will be 8-10 × Atlas and that they expect to gain factors from the architecture which increases the performance on real programs in high level languages (Fortran, for example) to 12-15 × Atlas. Evidence so far, based on hand-simulation of the compilation of Fortran loops, suggests that the lower figure is the more realistic. No machine has been built yet; the first proper engineered prototype should be switched on in November 1971 and the first production machine available for delivery in April 1973. These are ICL's dates. They have quoted us £2.4M for a large installation with 2 Mb of main store (250 ns) and 1,000 Mb backing store on fast discs, with delivery in April 1973. This would be for an independent machine. ICL are formulating a proposal for a P4 system, with minimal peripherals, linked to our 1906A and using the latter for all input and output and job organisation. They expect to complete this during the next two months.

CDC 7600

This machine, developed from the 6600 of which about 100 are installed, has now been in existence for two to three years and is thoroughly proved. It is much more of a conventional order code machine than P4 or STAR, with a power of 20-25 × Atlas. The Computer Board has financed a joint project between ICL and CDC for the Manchester Regional Centre, in which a 7600 processor and store will be linked to the existing 1906A. This is due to be in operation at the end of 1972. The Board is considering the case for a second 7600, this for the London University Centre (which now has a 6600) and possibly serving all the universities in the south-east.

A suitable configuration could cost about £3-3½M and could be delivered in 18 months. The Manchester system has cost about £2.2M for the CDC equipment and the link to the 1906A, but is a small system.

CDC STAR

The name means STring and ARray processor, and the machine has very novel architecture. It has what one might call a conventional order code similar to the 7600 and on that score is of comparable speed. But in addition it has a set of very fast vector operations which are such that if a vector C is obtained by elementary operations on vectors A and B (eg, C. = A + B, i = 1 to n), the successive individual components are formed in times which average about 36ns - about 30M a second, which is about 100 × Atlas. However, the conventional operations are significantly slower - an estimate of 160ns seems reasonable, though 80ns can be achieved - and there are overheads in starting up a vector operation and possible storage problems in retaining intermediate results. The vector operations are clearly a potential source of great power but a lot of detailed study is needed to see how fully this could be realised in a workload such as ours. Los Alamos, who have two 7600s, as well as three 6600s, have been studying the machine and have arrived, perhaps rather tentatively, at an estimate of l½-2½ times the 7600: say, 30-50 × Atlas; this is in a context of very large-scale scientific projects with many very long production runs. On the other hand a verbal statement from Livermore gives the machine a potential of about four times the 7600, say 80-100 × Atlas, if properly exploited.

CDC began development of STAR in 1966. Several hardware models have been made and the first production model is running in the development laboratory at Arden Hills near Minneapolis and is due to be shipped to Livermore in November 1971. A second production machine is being built. CDC have also engineered the final hardware model and are offering it as STAR-1B, with the same architecture but raw hardware speed about the same as a 360/50. This could be used by a STAR-100 customer to get advance experience of the machine's special capabilities and for software development. CDC have offered us a system for £4M, with delivery of a STAR-1B in April 1973, for development work, followed by the STAR-100 in April 1974.

We have a detailed paper on STAR written by Dr P E Bryant of the laboratory after a visit to CDC in America in May 1971.

SUMMARY

It is clear that one can get more computing power, more cheaply and with greater continuity, by buying one of the big CDC machines than from the ICL machine. The cost per Atlas power of the P4, 7600 and STAR seem to be, very approximately, £200,000, £150,000 and £100,000 respectively, for capital costs of about £2½M, £3M and £4M respectively.

The 7600 - which may be taken as representative of the conventional large machine - offers the most straightforward and least challenging solution. My own view is that it is less to be recommended than the others for this reason and because access to this machine is being made available through Manchester and possibly London also within the next l½ to 2 years.

The CDC STAR and the ICL P4 both have novel architectures. So far as is known, STAR is the most powerful machine for which production is planned - although it is known that CDC are designing a faster successor to the 7600 - and the exploitation of the vector orders offers great scope for development of new methods of attack on large scale problems. The ICL New Range has the built-in stack and uses associative stores to reduce the number of red-tape operations, provides very good store control and is planned to permit easy and flexible development of a large installation. The P4 is clearly a long way below STAR in power and is in a much earlier state of development, and we do not know if ICL have any plans for producing a faster processor in the range. It is, however, a British product, the only advanced machine being developed anywhere outside America, and the New Range is the only visible possibility for a new generation of British computers. The Committee may feel that this is an important factor in the choice of the Laboratory's major equipment. At the same time, the two companies, ICL and CDC, are collaborating in various ways, for example in the setting-up of the ICL-CDC-CII (France) joint company, and the Committee may feel that there should be a case for proposing a joint project involving STAR and a New Range processor. I should be grateful if the Committee would discuss this situation.

APPENDIX B

Minutes of the Atlas Computer Committee of 16 September 1971

These minutes are available as a separate document in full.

⇑ Top of page
© Chilton Computing and UKRI Science and Technology Facilities Council webmaster@chilton-computing.org.uk
Our thanks to UKRI Science and Technology Facilities Council for hosting this site