The role of the Atlas Laboratory was discussed in paper SRC 72-71 presented to the Council on 15 December 1971, copies of which have been sent to all members of the Committee. As members will know, the Council accepted the main proposals of the paper and in particular agreed that the Laboratory should specialise in the provision of computational support for large-scale research projects which had the scientific approval of the relevant SRC subject committees. We are now engaged in implementing the agreed new policy, with every indication that this will prove of great benefit to the scientific community and that this is recognised by the scientists. The Council paper discussed briefly the level of computing power likely to be needed in the Laboratory in the future. The present paper invites the Committee to take up the discussion and gives the Laboratory's views.
Like all SRC laboratories the Atlas Laboratory has to estimate the legitimate future demands which will be made on it and to plan ahead so that its resources will meet this demand. Its key resource is its principal computer, which represents a considerable investment in money and professional skills and must have long useful life, probably at least seven years: Atlas, for example, will have given nine years of invaluable service by the time it is closed down in 1973. The process of choosing, acquiring and installing a major computer and bringing it into effective service is unlikely to take less than two years, so we have the problem of deciding on a system which will remain viable at least until the early 1980's.
The main demands on the Laboratory come now, and so far as one can foresee will continue to come, from research workers in the British universities. The Computer Board provides for the needs of the general university computing services and, put very crudely, its plans will bring the average level or installed power up to about the equivalent of one Atlas for each 5,000 students. This is certainly as good as the average for America, but will be taken up almost entirely by the general needs of the undergraduate and postgraduate student needs, as is the case now. There are big and exceptionally well-equipped laboratories in America - for example, the AEC laboratories - to which advanced research workers can turn for computing services on a large scale, but in Britain there is only the Atlas Laboratory. Thus the choice of our next machine will have a profound effect on what computer-dependent science can be done in Britain from 1975 to 1980 or beyond.
The Scope of the digital computer seems to be literally unlimited. Its powers, now well recognised as going far beyond mere arithmetical calculation, have been applied in every field of knowledge and activity and the history of the last ten years or so has established its fundamental importance in science and technology. A great deal has been written on this and there is a lot of valuable and interesting material in the following recent references;
There does not seem to be any disagreement over the value of the computer as a weapon of attack upon problems in all fields of science or of the need for research workers to have access to powerful facilities. But it is exceedingly difficult to make an assessment of the level of power which should be provided. There is no doubt whatever that if one simply granted requests, any computing power which the Atlas Laboratory, for example, were to offer would be taken up. The problem is to predict the needs of good science - meaning in effect the needs of future projects which reach the standards required for SRC support. Obviously one cannot hope to make such a prediction with anything approaching precision, because one is dealing with the whole field of science and cannot say now what will be the important lines of research justifying large amounts of computing power in, say, 1978. One can only hope to get an indication of the minimum provision required, and the following observations and arguments are directed towards this.
The general tenor of the preceding statements is that one can see already a need for computational support for good quality science beyond what the universities themselves can meet, equivalent to about 10 to 15 times the power of a full-time Atlas. This then seems to be the absolute minimum level of computing power which can be justified for the Atlas Laboratory. At present, the Laboratory has about 6-7 times Atlas, made up of 4A-5A from the RHEL 370/195 and about 2A from its own 1906A. The demands already being made on the 370/195 by the nuclear physics community show that there will be very strong pressure within the reasonably near future to have the whole of the time on that machine; it does not seem realistic for the Atlas Laboratory to rely on its present allowance of 20 hours a week continuing indefinitely. The Laboratory should therefore be looking for a new machine which would provide for the whole of its requirements from about 1975 to about 1982. We have seen that a machine in the 10-15 Atlas class would be saturated very quickly, so to remain viable over so long a period it should be substantially more powerful than this, or at least capable of substantial up-grading.
Until very recently we should have said that only four machines need be considered: the ICL (New Range) P4, the IBM 370/195,the CDC 7600 and the CDC STAR. We have made very detailed assessments of these machines and reported them in the paper Proposal for a New Computer Installation at the Atlas Computer Laboratory - Part II: Machine Evaluations which is submitted to the Committee under cover of this paper. But in late August we had an approach from Texas Instruments and at the end of the month a presentation from them of their Advanced Scientific Computer (the ASC), a machine in some ways comparable to STAR. We do not have as much information on this machine as on the others and have not had time or opportunity to study it in such detail. We say as much as we feel able at this stage in an Addendum sent with the main paper.
It is extraordinarily difficult to give an objective assessment of so complex an entity as a large modern digital computer and we feel that our study goes as far as one can towards this goal: A very skeletal summary is given in Table I.
The Laboratory has come to the following conclusions (given in much greater detail in the above mentioned proposal) as a result of this study:
The view in the Laboratory is that STAR is the machine to go for. It would secure the computing resources of the SRC and enable the Council to play a leading role, on a world scale, in the development and use of computational attack on scientific problems. Because of its novel architecture it would also stimulate a lot of computing science interest. The drawback, of course, is that it is expensive and American. CDC have made us a formal offer for the configuration which we consider necessary; this includes various concessions, in particular a substantial allowance for software development by ourselves. The net price as shown in the attached paper on Machine Evaluations is just under £4M, with an annual maintenance charge of £147K. These sterling prices were given when the exchange rate was £1 = $2.4. CDC have now told us that they must protect themselves against any possible serious devaluation of the £ and would therefore now quote the price as $10.3M. They expect to be able to hold this price until about June of 1973.
In the end it is a question of the level of computational support which the Council is willing to provide for science. The Laboratory itself cannot assess the scientific merits of lines of research and development which are now, or in the future may be, making large scale demands for computing and needs the Committee's guidance here. But given a statement of the level of support which is required, the Laboratory can then say what equipment would be necessary and what would be the cost.
Computer | Delivery | Cost £M |
Performance × 1906A |
Cost per 1906A Power £K |
Notes |
---|---|---|---|---|---|
ICL P4 | 1975 | 3.5 | 8 | 430 | This refers to a 2-processor system. A 1-processor system could be delivered in 1974, would cost £2.5M and would give 4-5 × 1906A performance. |
IBM 370/195 | 1973 | 3.6 | 10-15 | 300 | |
CDC 7600 | 1972 | 3.0 | 12-15 | 240 | |
CDC STAR | 1975 | 4.0 | 12-150 | 26 | The very wide range of performance is due to the varying extent to which the fast vector orders can be exploited. |
ICL 1906A | 1971 | 1.1 | 1 | 1,110 |
User | University | Hours | Minutes |
---|---|---|---|
Chemistry | |||
Dr D T Clark | Durham | 51 | 7 |
Dr C Thomson | St Andrews | 46 | 5 |
Dr M H Palmer | Edinburgh | 26 | 48 |
Professor R McWeeny | Sheffield | 22 | 42 |
Professor G A Sim | Glasgow | 17 | 11 |
Dr W G Richards | Oxford | 11 | 28 |
Professor P G Perkins | Strathclyde | 6 | 1 |
Dr I H Hillier | Manchester | 4 | 7 |
Dr G Gerratt | Bristol | 2 | 38 |
Physics | |||
Dr K Singer | Royal Holloway | 78 | 45 |
Dr M J L Sangster | Reading | 10 | 41 |
Professor R J Hockney | Reading | 10 | 9 |
Dr M Petravic and Dr G Kuo-Petravic |
Oxford | 8 | 5 |
Professor J Powles | Kent | 6 | 14 |
Durham | 1 | 37 | |
Mechanical Engineering | |||
Dr R D Henshall | Nottingham | 4 | 52 |
Seismology | |||
Dr E P Arnold | Edinburgh | 3 | 26 |
Space Satellite Data Processing | |||
Miss B Stokoe | ACL (S-68) | 2 | 53 |
Dr R Dalziel | RSRS (IUE) | 0 | 2 |
This paper invites the Science Board to consider proposals concerning the computing service provided by the Laboratory to its users and the Laboratory's own programme of research and development work. The Atlas Committee discussed these proposals on 11 November and approved them in general terms; the Computer Board was invited to comment and gave its support at a meeting on 9 November. I have also discussed them with the Chairman of the Council and he too has indicated general approval. I incorporate Sir Brian's and the Committee's views into the paper, and give the relevant Committee Minute at Appendix A.
It may be helpful if I give the Board a brief summary of our present position. We have been giving a very general computing service to university research workers on Atlas since October 1964; this takes up about 70% of the machine's time, with the remainder split fairly evenly between work for government organisations (eg other Research Councils), who pay for all work done, and our own internal work. We take delivery of an ICL 1906A in June/July 1971 which will be installed in a new machine block now being built and will be operated in parallel with Atlas; this should add at least twice the computing power of Atlas and because of the large amounts of both core and backing store should give particular advantages in the running of large programs. We contemplate continuing Atlas until 1973 and replacing it then with a machine of at least 10 times its power, but otherwise unspecified; there is a provisional entry of £2.5M in the Forward Look for this. A very successful multi-access system has been developed, using an SDS Sigma-2 computer as satellite to Atlas, which could be transferred quite easily to a new machine given certain hardware provision. The Laboratory has the only piece of equipment (the SD 4020} available generally to university people for producing computer output on microfilm, and is in the process of installing a powerful interactive graphics system, in which a PDP-l5 computer (already installed and working) and a VT-15 display (due in December 1970) will be linked to the 1906A. In addition to the machinery the Laboratory has these considerable assets: with the new machine block, excellent and flexible accommodation for major computers, which will take two large machines simultaneously; very great expertise in computer science using that term in its broadest sense - and in the running of a big computing service; and exceptionally good and extensive contacts with the university world. The Council has in the Laboratory a very powerful weapon at its disposal.
When the Laboratory started operations in 1964 it was appropriate that it should offer a general service, because the level of computing equipment and expertise in universities was then very low. But since then there has been the Flowers Report, followed by the setting up of the Computer Board and the consequential great improvement in the universities' computational circumstances. Thus the computational environment is now very different. The Laboratory's administrative position is very different also; it now belongs to the Science Research Council which did not exist in 1964 and which was set up with the responsibility for supporting scientific research in British universities. The responsibility for providing them with general-purpose computing services has been given to the Computer Board. Thus I feel that on all counts it is necessary to take a fresh look at the Laboratory's role. My proposal is that, starting with the 1906A, we should move gradually from the present general service to one in which the major projects came to us through the SRC Boards and Committees; and that we should actively encourage scientists to formulate large-scale, long-term computing projects and should guarantee machine time and supporting services to those which had been approved on scientific grounds. The advantages of this would be first that the Laboratory would be more closely involved in the scientific aims and policies of the Council and could be used in a positive way in support of these; and second that by giving guarantees it could make it possible for the leader of a research project to plan the work with the assurance of getting the computational resources it needed, possibly over a long period.
It would not be practical to channel through the subject committees all the large number of applications which make up the present Atlas work-load; I am thinking in terms of a few tens of really big applications annually. The true cost of running the 1906A will probably be about £150 an hour, so a project needing 100 hours on this machine will be costing about £15,000 in machine time alone; the Council does not give a grant of this size unless the scientific value of the work has been vouched for by one of its subject committees, and there is no reason why computing should be exempted from the normal scrutiny. The guarantee is an essential feature: grants are made for a stated period and an applicant for computing time must know how much he can get, and at what rate (eg hours per day or per week) in order to plan his work.
I do not think that anyone doubts the existence of very large computational needs in worthwhile scientific work. Computing is clearly becoming increasingly important in all fields of research. However, to be a little more specific, given in Appendix B is a summary I have made of the needs stated to Dr K V Roberts's Working Party on Computational Physics, from which it is clear that the need for large amounts of storage is emphasised as much as is the need for fast processing. I have started to follow up some of these lines by talking with individual scientists, and I and my colleagues will do a good deal more of this during the next few months; in particular, I want to look into the actual and potential needs in engineering. I might just say here that there seem to be well-defined needs for computing projects on a very large scale in theoretical chemistry and in what one may refer to as theory of materials. One explicit example of the latter is an application to the SRC from Dr K Singer of Royal Holloway College for a very extensive study of topics in the physics of amorphous materials, a field which the Council wishes to encourage. We shall be meeting members of Professor Cruickshank's Department (UMIST) in early December to discuss their needs for molecular structure calculations. The Board may be interested to know that discussions are going on in America about a possible National Centre for Theoretical Chemistry, which would have a very powerful computer ("in the $5-l0M range!) dedicated to the subject. And finally, we are already committed to doing the data-processing for the UK-4 space satellite, which will probably need 5 to 10 hours a week on the 1906A from October and running for about two years.
The Atlas Committee, whilst supporting this proposal in general terms, said that it would not wish to see it applied in such a way as to restrict the use of the Laboratory to people working in the fields selected by the SRC for special support, and this is certainly not my intention; I would expect that good science would be supported, whatever the field. I would expect also to retain discretion to use any time left after all guarantees had been met for work not carrying a guarantee, coming from universities or from the Laboratory itself. In fact it would be essential to have such work available to give a good job-mix and to fall back on whenever one of the large programs failed.
It might often be desirable, or even necessary, for someone carrying out a large-long-term project to spend a lot of time in the Laboratory or to have one or more assistants doing so; this happens already and we designed the building and provided services with the possibility in mind. With the proposed concentration on large projects there could be an increased need and a closer involvement of the long-term visiting users with the Laboratory staff. This is something which Sir Brian Flowers would welcome; his view of the SRC is a body which works in partnership with universities, rather than providing passive support.
The proposal represents a considerable change from our present method of working. I should prefer to let the Atlas service continue as it is now, because many people are relying on it and a lot of the work done on the machine could not be easily transferred elsewhere. But we are free to choose how we shall run the 1906A and any subsequent machine- actually, a very important difference between this Laboratory and a university computer installation, which is subject to pressures which do not act on us - and we have a most valuable opportunity to introduce a new kind of service.
I have mentioned in this Paper that we have two pieces of specialized ancillary equipment: the SD 4020 microfilm recorder and the interactive graphics system. These are specialized in function but can be of value to a very wide range of users - the SD 4020 is of universal use. They are both expensive, costing around £100,000 each, and need considerable software and operational support to make them easily and efficiently useable. Thus they are very much the kind of thing to be put at a professionally-run centre. I feel that part of our policy should be to install ancillary equipment, of which the above are examples, which meets special types of need and which is either too costly or needs too much support to be installed in more than a very few places. It would of course be up to the Laboratory to identify the need and make the case to the Council. It seems that there may be needs for an input/output device for information on microfilm and for a really flexible text reader.
It is not possible to run a big and varied computing service without doing a good deal of software development. This was recognised from the beginning in our case by the setting up of the Programming Group which has produced, among other things, the Fortran and Algol systems, the multi-access system and the software for the SD 4020 microfilm recorder, and has adapted for Atlas the X-Ray 63 package of programs for crystal-structure calculations. We have found it very valuable, for thoroughly practical purposes, to have available such sophisticated basic software as a compiler compiler and a compiler for a string-processing language (SNOBOL). The aim of this internal work has always been to produce advanced software tools which were of value to large numbers of users; some, such as the microfilm software, could be valuable to virtually every user, others, such as the crystallographic package, could meet the needs of a restricted but still extensive class of users. An important point is that the product is never tailored to an individual need but must incorporate a great deal of generality. I feel that this forms an extremely important part of the Laboratory's support of science, an essential complement to the straightforward computing service, and I should like to see it put on a more formal basis. I propose that the Laboratory itself should submit applications for research and development projects and that these, if approved, should be given guarantees and subjected to the same conditions (eg reporting of progress) as any other applications. As with the service work, I should expect to retain discretion to undertake small-scale work without going through this formal machinery.
This side of the laboratory's work could form part of the SRC's support for computer science and could be related to the work of the Computing Science Committee. We are very well placed to give this support, having on the one hand both expertise and equipment and on the other the discipline imposed by a large-scale service, which puts anything we do to a severely practical test.
Sir Brian Flowers supported this part of the proposal also, feeling that a sound programme of internal work could significantly improve the service which the Laboratory could give to its users. He felt that it should not take up more than about 20% of the Laboratory's effort.
The 1906A has always been regarded as a first step in the re-equipping of the Laboratory. Our large configuration, with the proposed concentration on large-scale work, should give us at least the expected 2½ × Atlas computing power and I am privately expecting to get more from it; but all the evidence we have of the potential demand shows that much more power will be needed to give an effective service to scientists with really big problems. I mentioned earlier our hope of replacing Atlas in 1973 with a machine of at least 10 times its power: these figures are very imprecise - the performance of a computer is something one is hard put to define and varies greatly with details of the configuration, the operating system, the compilers used and the workload - but I consider this the minimum requirement. The Working Party on Computational Physics (see Appendix B) has, in effect, assessed the need at around 20 × Atlas. My aim is to get a sufficiently realistic assessment of the need for me to be able to put a specific proposal to the Council well before the end of 1971.
I must mention the Rutherford Laboratory 360/195, which will be installed in late 1971 and on which we can expect to get 10-20% of the available time for an initial period. We intend to use this time for the very biggest projects, which really do need a machine of this size and power.
I have already mentioned the need for sophisticated ancillary equipment; the order of expenditure I think likely is what is shown in the Forward Look.
For staff, the Forward Look shows an increase from the present 118 to a ceiling of 150, when two major machines are fully manned. The 150 is made up of 80 for Operations, 20 for Administration (which includes all the general services - clerical, typing, cleaning, transport etc) and 50 for the Programming and Support Groups and research posts combined (and myself and my secretary). Whilst I should not want to say now that we shall never ask to go beyond 150 I do not see any need for a big increase over this; but I can see that we might want to change the distribution of the 50 - which includes all the Scientific Officer class staff - amongst the various activities and I should like to have this regarded as a unit.
What I am saying in this paper is that I should like to see the Laboratory more closely integrated into the Council's scientific programmes, and that my proposals are to my mind the best way to bring this about. I believe that they would result in a better use of our resources in support of science. Also, that they take advantage of our special position as an SRC laboratory that, whilst our purpose is to support university research workers, we are independent of the universities themselves and free from the pressures and restrictions to which a university computer laboratory is subject.
1. The Working Party met on eleven occasions between November 1969 and May 1970. Invitations to submit papers or otherwise comment were sent to all Physics Departments. Brief comments were received from four, and twenty-one papers were submitted, including those written by members of the Working Party and excluding those dealing with organisation and reporting. These concerned the following branches of physical science:
There were also papers, among the twenty-one, on numerical analysis, computer software and the organisation and distribution of computing facilities.
2. Summaries of the papers are given below, so far as they give information about the level and nature of the computational demands of the various subjects. The general conclusions drawn by the Working Party were:
3. The needs described in the papers could be met efficiently by a dedicated machine of 10-20 times the computing power of Atlas, at least 1Mb fast core store, very large backing store and extensive peripheral equipment including graphical recorders and interactive visual display. The cost would be around £3M. The Working Party recommends that such a machine should be ordered as soon as possible, and installed at the Atlas Laboratory, with funds provided to allow data links to be installed to selected centres.
4. Much work on software development will have to be done to make this central machine easily available to workers in different fields who have different experience of, and approaches to, programming; and especially to enable them to develop programs on local university machines and make the production runs on the central machine. The Working Party recommends that machinery should be set up for the co-ordination of software development and establishing of standards, with the object of increasing the availability and transferability of large programs. This would best be done under the aegis of the Computer Board. Equally the Working Party feels that a computer network will become essential and that a professional organisation will have to be set up to specify this.