Brief details of a proposal to site an Atlas computer at the Rutherford Laboratory were given in paper NI/60/14. The following further notes may help in the discussion on January 30th.
I suggest that a Computer Committee would need to be formed to advise the Board on all matters concerning the Institute's interest in the computer. It would include several Board members together with computer specialists and users of such standing that the Board and the university users could be expected to find their recommendations acceptable. I suggest that the man responsible for discharging the Institute's function on the computer should report directly to the Computer Committee. I believe that both AEA and NIRNS staff would have to be involved in operating the computer; details of the collaboration between the Authority and the Institute in this matter would need to be worked out.
You asked me this afternoon to advise on the position of the National Institute for Research in Nuclear Science in relation to the proposal to install their Atlas computer. The computer will be used, at first, only for a limited part of the available time on nuclear research and related matters. The intention is that other users, particularly universities, shall be allocated time on the machine and you are anxious first of all to determine whether it is within the Institute's competence to allow use of the computer in this way. Secondly, you would like to know whether the matter is at all affected by whether a charge is made for these non~nuclear uses or not.
I pointed out that paragraph (d) of the Institute's Charter is the only paragraph in Article 4 in which no express reference is made to nuclear science. It seems to me that if the National Institute has decided under paragraph (b) to provide a computer, it would be open to the Institute to permit the computer facilities to be used by scientists for scientific purposes falling outside the field of nuclear science. As an example, I think that if the Weapons Group of the Authority wished to use the computer for some research programme into ballistics and the Institute decided that such use of their facilities was appropriate, no question could be raised as to the vires of such use of the facilities. I do not think the word appropriate is to be construed in the narrow sense of meaning appropriate only in connection with research into nuclear science, but is to be read rather as meaning appropriate in all the circumstances of the case within the Institute's general discretion. Obviously one of the points which might move the Institute to consider some use to be appropriate would be the fact that it would be better for the computer and its operating staff to be fully employed rather than to be idle, they might also consider it to be appropriate that scientists generally should be permitted and encouraged to acquaint themselves with the use of computers in the scientific world. This would presumably ultimately lead to an improvement in the range and capability of computers, which in itself would be of advantage to nuclear science.
It does not seem to me material whether or not the Institute decides in any particular case of a scientific use to make a charge for the facilities they are providing. It might be argued, however, that Article 5 of the Charter (which provides that all moneys and property received by the Institute shall be applied solely towards the promotion of the objects of the Institute) indicates that any non~nuclear scientific user ought not to be subsidised by the Institute and that therefore some charge for the facilities provided to non~nuclear scientific users ought to be made.
I notice in Turnbull's paper there is a proposal to let the machine out on hire to industrial firms for non-scientific purposes at commercial prices. A user of this kind would not, of course, be covered by the express provisions of the Charter, but I think that the Institute would not, in practice be challenged if it decided in its discretion to permit such user, without prejudice to scientific interests. The Treasury would not be at all likely to object, as their grant to the Institute would be abated by the commercial receipts and I do not foresee any possibility whatsoever of a private individual such as a disgruntled computer manufacturer complaining to the Courts about this user and being granted an injunction to restrain the Institute from continuing to let out the spare time capacity.
As I explained, it is not my job to advise the Office and in particular I am not (anyhow officially) the legal adviser to the National Institute. It was therefore agreed that I should send a copy of this letter to Woodhouse at the Treasury Solicitor's Office and ask him to get in touch with you if he disagrees with anything that I have said.
The National Institute for Research in Nuclear Science have considered the proposal that an Atlas computer should be installed at Harwell under their control and management. I am instructed by the Chairman to let you know that the Institute are prepared to accept this commitment. In recording their decision the Institute added that they considered that they should not charge the universities for use of the computer, either for nuclear science or for any other use.
The Institute were by no means convinced that the proper place for a university computer was at a centre remote from a university, but they recognised that there were special considerations in this case, and they took note in particular of the following points, which were put to them as being the views of the Minister's Working Party.
The provisional specification drawn up for the Harwell Atlas is for a large-scale installation costing a little over £2.5m. If this is approved the computing laboratory in which it is to be sited will have one of the most powerful machines in the world - possibly even, the most powerful, for there are indications that Atlas may turn out to be a better machine than its only rival, the IBM Stretch. Thus, this laboratory will have a most serious responsibility, that of ensuring that the best possible use is made of a major scientific asset; and as the fields of application of mathematics have been widening and diversifying with rapidly increasing speed since the digital computer became an accepted tool - that is, over the past 5 years - we cannot predict what will be the most important problems to be put on to the machine in 5 years time; therefore the laboratory must be prepared for anything. I want to give here my views on the way the laboratory must be staffed if this challenge is to be met.
In the collaborative scheme, which seems to have been accepted, the laboratory will have to provide various services for a variety of customers, and will belong to Universities, Research Institutes, Government departments and the AEA; these are:
This will require the following staff groups:
I feel very strongly that all these people should be together in the computer building; computation brings one into contact with all kinds of people and a balanced group is necessary to give the right climate. I am strongly opposed to the suggestion, which has been hinted at, that the laboratory should confine itself to operating the machine - items 1 and 2, that is - and leave everything else to the users; this would be disasterous to the success of the project and is equivalent to a suggestion that the new synchrotron should be run by a group made up of the maintenance engineers and a few clerks to keep the books.
The size of the group, excluding the maintenance and operator teams, which can be settled on straightforward considerations - will need some discussion; it should not be too small, for there must be plenty of scope for interplay of minds; and a fair rate of flow through the group should be encouraged by having several posts for Fellows or Research Associates. I would suggest something around 30-40. A very much more serious subject for discussion is the future of the Harwell computing group; does this become part of the new organisation or does it become just one of the customers? The first alternative leads to difficult problems of control and responsibility, the second will cause a serious decline in the health and status of the group. The question will have to be resolved quickly, for there is a very great deal of work to be done, in all fields, in setting up the installation and this will not go well unless the final outcome is clear from the start. I favour the first course, with some very hard thinking to decide the form of organisation which will meet the undoubted difficulties.
First of all, I cannot see any good reason for its forming part of the Harwell Theoretical Physics Division; I think my group has gained greatly in the past from this association even though we have led a fairly independent existence, but with a new building (outside the fence) and much wider responsibilities, the links in future will be so tenuous as to be meaningless. So let us consider it as a Division, with a number of Groups.
The laboratory will have these responsibilities:-
This means that one wants this kind of organisation:
My estimate for the size and composition of the staff are these:
In addition there should be two secretaries (one for the Division Head, one for the Group Leaders) and 2 typists for mathematical and general typing, also a DAO - who might, however, be shared with another Division or a part of the Institute. The totals are, for 3-shift operation:
Type | Division Head |
Machine Group |
Computing Service |
Mathematics | Totals |
---|---|---|---|---|---|
Scientific | 1 | 5 | 23 - 27 | 9 - 13 | 38 - 46 |
Clerical and admin | 5 | 5 | |||
Operator | 12 - 18 | 12 - 18 | |||
Totals | 6 | 17 - 23 | 23 - 27 | 9 - 13 | 55 - 69 |
Visitors | 8 |
For comparison, the present complement of the Harwell Computing Group is 26 Scientific + 10 Operator staff and 1 secretary.
The question of formal control of the laboratory staff presents a difficult problem. The laboratory is intended to serve the AEA, the National Institute, the Universities and some Government departments; at the time of writing, it is being said that no charge is to be made to Universities for time on the machine. The AEA research Group will be dependent on the new laboratory for its computing service and must therefore have its needs and rights very carefully protected. To get a rough estimate of the cost of running the installation, let us suppose that the computer, costing £2.5 million, is written off in 5 years - a pessimistic view; that maintenance costs £150,000 per year; and that the staff costs are equivalent to 50 people at £5000 per year each. Other costs - depreciation of the building, power - are likely to be small in comparison. We get £500,000 per year; this can be exhibited as £180 per hour for time on the computer, for 3-shift working at 5000 hours per year.
There seems to be three ways of organising things:
These are some sketchy statements with a lot of difficulties glossed over; I would like to discuss the problem a great deal more.
You enquired whether the Authority would be willing to enter into a scheme which would enable university people to have computing done by Authority machines, until such time as Atlas becomes operational.
The question came at an awkward time because we had just received official confirmation from the I.B.M. Company that Stretch would not be as powerful as we had believed when we signed our contract with them. We have now negotiated a new contract the outcome of which is satisfactory to us in that we can still get our important defence work done, and our civil work as well, at the same cost as we originally estimated. In addition, our minimum financial commitments have been reduced. One disadvantage, however, is that Stretch will be about three months late and will not be operating until May of next year. Because of this delay the 7090 will not leave Aldermaston until about June, but it should be running again at Risley by July.
Since Stretch will be slower than we had hoped, the Authority will be using more hours per week than was expected, and for this reason, and because of the three months delay, cannot offer as much computing service in aid of the universities as would have been the case but for the recent changes. We can, nevertheless, still make an offer, which might be attractive.
On Stretch, we could offer up to half a shift (i.e. twenty hours) per week, starting late July 1962 and continuing for about two years, by which time Atlas should be ready.
On the 7090, the maximum we could offer would be an average of one shift (i.e. forty hours) per week up to the end of 1962, and an average of half a shift (i.e. twenty hours) per week in 1963.
If these offers are accepted, the Authority will be running both machines to their maximum capacity.
While we believe that we could undertake on the Stretch and the 7090 as much work on behalf of the universities as stated above, we must, of course, reserve the right that if an emergency computing job for the Authority arises the university work will have to be deferred for a time. We would, nevertheless, do everything possible to carry out the university work to the limit of our offer such as, for example introducing some delay into those Authority computing jobs, which had no urgency.
From our experience on computing we would not be surprised if some of the university requests for work to be done on Stretch or the 7090 would not be better handled by a smaller computer, such as a Mercury, in which case they should certainly be done that way. There is a Mercury at Harwell and we think we could fit in a few jobs for the universities on this machine. (in due course, we expect to dispose of this machine, but we shall keep it at least until the Atlas is running).
In making the above proposals the following points have been taken into account:
An organisation will be needed to co-ordinate arrangements for getting the work done. In considering the organisation, three periods have to taken into account. First, the next two or three years, when university work will be done on Authority machines. Second, the next two or three years, when the Atlas will be installed and will be doing a good deal of university work. Third, a period when universities will be acquiring Atlas computers and the university load on the first one may decrease. In this third period, the Authority will in all probability have more work for the machine.
An organisation which commends itself to us for the first period would have as its Head a man now at Harwell. This man would have the general oversight of all computing, including Harwell work. He would take his broad directions on the work for the universities from the National Institute Computer Committee, and he would be responsible locally both to the Director of the Institute and the Director of Harwell. It would be his responsibility to see that the university work, approved and accepted by the Institute Computer Committee during the period before Atlas is available is done on Authority machines, or if necessary, on other machines.
The Director of Harwell would keep together his teams for programming and operating Mercury, but the Institute would set up a small team for advising and helping the university people to programme their problems. Where there is "debugging" of programmes to be done, the National Institute staff would give any advice or service required, consulting the primary customer if necessary.
When the Atlas arrives, the Head would become an Institute man, although if he were from Harwell, he would remain on the Authority's books, seconded to the Institute. Harwell would keep its own programming team but the operating staff would be transferred to the Institute. The Institute would also have its own programming team, in order to programme its own work, and advise users, other than the Authority.
If there is a third period, where the university work moves to university machines and the main user is the Authority, the machine might be transferred to the Authority, and the Authority would provide a service to other users. The plans being made are sufficiently flexible to be able to permit an adjustment along these lines, should this appear desirable at the time.
Obviously these suggestions will require the Directors of the National Institute and of Harwell to take action in the near future. Conversations have taken place between them and both agree with the suggestions made above. The suggestions now being made are also consistent with the proposals being made in a paper by Pickavance and me to the Board of the National Institute.
With regard to the financial arrangements on the use of Authority machines for university users, we shall have to rely on you to persuade the Treasury. We consider that the Authority should be reimbursed by the actual amount that the service costs us. In other words, we would not ask for standard commercial rates. A suggestion which might be acceptable to everybody is that our charges should be about the same as those made to Authority users; namely a pro rata payment for the direct cost of the machine including overheads. For your guidance these costs would be about £335 per hour on Stretch and £120 per hour on the 7090. (We understand unofficially that I.B.M. will be offering a 7090 commercial service by July of this year and that their rate will be £250 per hour. Our quotation is lower than this because we have a guaranteed large usage by the Authority, we spread the costs evenly and of course we shall not seek to make a profit at the expense of the universities).
We would not wish to quote a maximum time or rate for the use of the Mercury at Harwell, but I suggest that some financial provision might be made for such work in case it should arise. We would expect to charge on the same principle as for Stretch and the 7090. Financial provision of £10,000 for the 1962/63 financial year should be enough to test the market.
Since receiving your letter of July 25th we have reviewed our forecast of the use the Research Group would make of the N.I.R.N.S. Atlas computer.
The estimate of 25+ hours a week was based on a series of deliberations by the Authority Computer Policy Committee during 1959-60, and was reached about a year ago. It was a rough estimate, and it is difficult to be more precise now, but it is already clear that the estimate is very probably too low. We would now put the figure at somewhere between 30 and 40 hours a week. The principal areas of use are (a) reactor physics, (b) solid-state physics, (c) plasma physics (Culham Laboratory etc.), together with a variety of miscellaneous uses adding up to between 5 and 10 hours a week. As far as we can see none of these areas are likely to be affected to any material extent by cuts arising from national economic developments.
I presume you have asked the Rutherford Laboratory of N.I.R.N.S. for its own estimate of future computer requirements (the earlier estimate was about 8 hours a week on the Atlas), but, judging by experience at C.E.R.N., Berkeley (U.S.A.) for example, the Rutherford Lab. Orion will have heavy demands upon it, e.g. for bubble-chamber track analysis, and the use of Atlas will increase.
As I understand it, London University propose to spend initially about £1,550,000 on a basic configuration which they call Atlas I, and to add later further equipment at an additional cost of £450,000 to form Atlas II. I do not know what provision has been made in these figures for a contribution towards the development charge. If by any chance the N.I.R.N.S. Atlas is deferred, a larger proportion of the Ferranti development charge will presumably fall on the London Atlas. The costs I have quoted are for the machine alone, and do not include anything for building and services, but I understand these should not be large since some houses in Gordon Square are available and would need only modification and additions.
To finance the computer, London University budgeted for a £500,000 grant from the treasury through the UGC, £500,000 advance plus one quarter of the operating costs from an industrial concern in return for 1,250 hours a year use of the Atlas for five years, leaving £1,000,000 to be borrowed from the Midland Bank.
London estimated the computer operating costs to be £165,000 a year. I am sure this is too low, quite apart from the fact that no provision has been made for amortisation or overheads (building maintenance etc). On the figures given the industrial concern would be paying £100,000 + £41,000 = £141,000 a year for 1,250 hours use. At approximately £115 an hour this is a heavily subsidised rate. We estimate that the minimum inter Group paper charges that we could make for the use of the NIRNS Atlas (getting on for twice the size of the initial London one) would be £350 an hour on present prices, and commercial rates have been estimated at between £500 and £750 an hour. This means that a fair price for the London Atlas hire by an outside user would be not less than £250 an hour and probably appreciably more.
Since London University will have to sell time not used by themselves or the privileged user, they will not be able to permit other universities, eg Cambridge, to have time free of charge. Thus these universities would use the NIRNS Atlas according to the arrangements already agreed by which the use will be free to university, Thus if both projects go through, the loss to the NIRNS Atlas would be only the projected London University use, The present trends indicate that this loss will be made good by other users not long after the NIRNS Atlas is fully in commission.
I presume the Office of the Minister for Science will obtain the UGC views. As you will appreciate, some of the information I have given above is in confidence.
I have received from Dr Howlett a minute suggesting the type of performance test which we ought to impose upon the Atlas machine. His suggestions are as follows:-
1.1 A simple loop of floating-point additions:
Similar loops for subtraction, multiplications and division.
The same loops including (n,B) instructions alternating with the arithmetical instructions.
1.2 A loop to form the scalar product of two vectors.
1.3 A loop to evaluate a polynomial
The timing for these shall not exceed those computed from the figures of Col. C. of the tables in Dr Wilson's letter of May 19th.
The above programmes to be run whilst peripheral transfers are taking place; that is, transfers to or from any combinations of the following:
The loss in effective computing speed, due to the peripheral transfers, shall not exceed the figures given in Dr Wilson's letter of June 2nd.
A programme to invert a (large) matrix, using a specified method. This is to be run with and without peripheral transfers in process. The time for this programme will be computed in advance.
All these programmes will be written by Ferranti's in collaboration with A.E.R.E. and we will formally accept an agreed set. The precise method of timing will have to be agreed.
Will you please let us know whether you can agree with this suggestion fairly quickly or whether you will need time to study it and possibly to make suggestions for change.