Minutes of the eleventh meeting of the Computer Policy Committee held at 2.30p.m. Monday 7th January, 1963 at Charles II Street, London.
1. The Chairman announced changes in the Committee. Dr. Howlett had given up his post at A.E.R.E. to become head of the Atlas Computer Laboratory and could no longer represent A.E.R.E. on the Committee. Dr. Howlett would attend future meetings of the Committee as an ex officio member since a large part of the Authority's future computing needs were bound up with Atlas. The Chairman welcomed Dr. Curtis (A.E.R.E.) and Dr. Roberts (Culham) as full members of the Committee.
2. In paragraph 2 replace to run Stretch for 10 hours by to run Stretch for 16 hours and in paragraph 7 change the word cores to stores. Dr. Black reported that extra staff for the 7090 had been interviewed and offers sent: he hoped to operate the 7090 at 125 hours per week from 1st March, 1963. Dr. Corner said that AWRE had averaged 139 hours a week for six months. Dr. Black replied that since he was limited to four teams of operators, and allowing for staff holidays and machine failure, he could not guarantee more than 122 hours useful time per week. He would, of course, run for a greater number of hours when possible. The Chairman said that he had written to Mr. Turnbull and offered to the Universities up to an average of 20 hours per week on the Harwell Mercury until the time when Atlas became operational. Dr. Corner said that he disagreed with one conclusion of the last meeting, (which he was unable to attend). He thought it would be possible to offer Stretch time to the Universities. The Chairman asked him to bring this up later.
3. The Chairman said that the demands for computing in the next year seemed fairly clear. For subsequent years the estimates contained some uncertainty; it was being assumed that the demand would follow a rising curve, and it looked as though not enough computer time would be available. The cost of computing was becoming a substantial part of the total vote, and members would realise, that since the Authority's budget was constrained, if expenditure on computing were to rise then economies in some other field must be made. The Committee could not make the necessary overall judgement, the time had come to ask the various Groups Managements if they were prepared to pay for all the computing that was being estimated.
4. The Chairman pointed cut that the Stretch Contract would run out in 1964. There was no indication that we would not renew the contract, but it should be realised that any final decision rested with the Military and not with the A.E.A. Dr. Corner said. that the Director of A.W.R.E. had given him a statement for transmission to the Committee which read: - The Stretch contract is open-ended: there is no date specified at which it must end or be renegotiated. The earliest date at which the contract would permit the AEA to cease using Stretch is the 21st August 1964. I have no intention at present of suggesting to the A.E.A. that it should take advantage of this, and I see no reason why the C.P.C. should not assume that Stretch will still be at AWRE at the end of 1964. If the C.P.C. prefer to work in terms of financial years then I would be happy to make the same statement but applying to the end of March 1965, and on the question of the rent of Stretch - An attempt to negotiate a lower rent for 1964 would require A.E.A. to have, up its sleeve, the implied threat to change in September 1964 to another cheaper machine of equal power; such a machine may still be announced, with a delivery date as early as that, but if so it must be announced soon; certainly no such machine is yet known. If no such machine turns up soon we shall be in a weak position in asking for a reduction of 1964 rent. IBM policy is to charge the same rent to all customers and they will emphasise that their hands are tied. I see little prospect that all the other hirers of Stretch in 1964 will be prepared to fight hard for a reduction, and on Winfrith's investment in Stretch program - any proposal to change Stretch for some other machine, the loss of their invested programming tuning-up effort will certainly not be forgotten. This can be guaranteed, for AWRE will be in exactly the same trouble.
5. The Chairman said that he intended to seek managerial endorsement of computing needs for the various Groups; he would then present a paper to the A.E.A. summarising the position; he would also explain the Civil needs on Stretch to the Executive, pointing out that ample warning would be needed if the military proposed any changes.
6. Dr. Marshall presented his paper (CPC/P44). Fortran IV produced by I.B.M, had failed to come up to expectations. The consequences of this had seemed to be very serious but it had become apparent in the last few weeks that the S1 compiler would be improved. The new version of S1 (which is called S2) should compile a program in a twentieth of the time taken by S1. The gain in execution speed was not clear but it seemed safe to take one Stretch hour as equivalent to 2.5 hours on the 7090, and there was some hope of eventually putting this figure up to 3.0.
7. The Chairman said that Mr. Glennie of A.W.R.E. who was responsible for producing S1 and S2, was to be congratulated on a fine achievement. Dr. Marshall said it was expected that S2 would be ready in April 1963, however the Authority should urge IBM to continue helping Mr. Glennie. The Chairman thought that IBM, in their own interests, would wish to further S2. Dr. Corner said that this was not necessarily true. Users in America tended to use Stretch for very long production runs and they were not worried by compiler inefficiency. A formal letter of request to IBM (U.K.) might assist them in their dealings with IBM Head Office in New York. The Chairman asked Dr. Corner to sound out Mr. Hudson of IBM whether such an approach would be welcome and report back if necessary.
8. Dr, Marshall presented CPC/P45. He said that Harwell needs for machine time had risen because of increased estimates and because the Shielding Group (which belong to Winfrith but is situated at Harwell) had previously been omitted. In his opinion nothing could be settled until the Atlas question had been decided.
9. Mr. Mummery said that it had previously been assumed that Atlas would cover all possible needs but this now appeared to be untrue. Dr. Marshall pointed out that various things had changed. Firstly computing needs had risen more rapidly than expected. Secondly Atlas would arrive later than had been expected. Thirdly it had been assumed that Atlas and Stretch would be much faster: the detailed speeds of the machines had not changed so much as our assessment of what this meant to total running time.
10. Dr. Howlett was asked if the Authority would be allowed to have two shifts on Atlas. He thought this was possible in the early days of Atlas, but improbable later on. Dr. Curtis pointed out that this was precisely the opposite of what would be required. On payment for time on Atlas, Dr. Howlett said that universities would receive free time, Government departments and the Authority would pay cost price, any others would be charged a commercial rate.
11. Mr. Hudspith was asked what obligation we had to use Atlas and what time could we be certain of. He replied that NIRNS obtained its money through a subhead of the Authority vote; we were obliged to use Atlas as much as possible for our computing, this saved money for the Authority as a whole, but we could not demand a firm allocation. Dr. Curtis commented that this could mean that work had to be switched from other machines and this would cause great inconvenience. Dr. Marshall thought that we could rely on one shift and that we were, therefore, only obliged to use one shift. The Chairman said that the Atlas Committee was seeking from all prospective users an estimate of their needs and the allocation of time on Atlas would become clearer by the summer.
12. Dr. Marshall suggested that if only one shift on Atlas was available then Harwell should have it all, Culham should use Stretch; if there were two shifts then Harwell and Culham should have it: Winfrith should not plan to use Atlas. Mr. Mummery said that he was not happy about the demands quoted in CPC/P44, these could be quite wrong. If no restriction was placed on machine use then demand would exceed supply, but financial restrictions could reverse the situation. He also felt that some possibilities had been overlooked. If Winfrith was going to spend an additional £150,000 it would be better off buying a machine of its own, thus saving £47,000 a year on data links. Dr. Marshall said this point had been covered. A new machine if ordered now would not arrive before 1965 by which time the £150,000 would have been spent. Mr. Mummery had no definite evidence on this point but he thought a machine could be delivered before 1965. Dr. Corner pointed out that Farnborough were about to order a KDF9 and the delivery date specified was April 1965: Dr. Black confirmed this.
13. Dr. Roberts thought there was no question about the Culham estimates. Culham had decided to allocate a fixed portion of its budget to computing; as computing costs per hour came down, the requirements for machine hours would rise. Mr. Mummery thought that if everyone did this then demands would be inflated: the current estimates were inconsistent, some estimates were rising, some were stationary. Dr, Marshall pointed out that the stationary figures belonged to Risley who had their own machine: Harwell estimates showed a rise because they expected a machine on-site to stimulate demand.
14. The Chairman said that he believed the 7094 issue was now clear provided present demands were confirmed by the Group Management Board. By 1964 the needs of the Northern Groups and Winfrith would saturate the 7090. By 1965 we would have paid an additional sum of about £150,000 to IBM for the increased use of Stretch if the conversion was not carried out. The difference between this and the expenditure required for the 7094 conversion was worth paying because the 7094 would fully recoup the costs in 65/66 and after: moreover the conversion represented a permanent gain in computing capacity. Mr. Hudspith said that it would be difficult to convince the Treasury that a 7094 was needed unless Atlas was fully used. The Chairman believed that 1 to 1.5 shifts on Atlas would meet our commitments in that direction and other users would want the rest. Dr. Marshall pointed out that the total Atlas usage was not really relevant to the argument. We should start using Atlas in April 1965, it would take at least six months to transfer all programs from Stretch to Atlas and by this date £280,000 would have been spent on hiring extra Stretch time. Mr. Iliffe said that if the figures for machine time in April 1966 were accepted then at this date Winfrith would begin to increase its usage of Stretch, even with a 7094 at Risley. If a new computer were to be considered in about two years time, the Treasury case for it might have been weakened by the 7094 conversion. As the financial difference between Stretch usage and 7094 conversion was small, he thought the 7094 conversion was the less attractive choice.
16. Mr. Mummery also thought that we should not rule out a new machine. The saving on hire of Stretch plus the saving of £40,000 a year on data links provided the basis for a sound financial case. The Chairman said that there was no evidence that a new machine could be obtained in less than two years and the capital costs were substantially greater. However, Mr. Mummery was free to make enquiries after the meeting.
17. The Chairman said that he proposed to write to Mr. Moore and Dr. Vick outlining the situation: if Groups required the machine time which had been estimated then the Authority would have to spend an extra £0.25M either on Stretch or on a 7094 conversion; this Committee thought that the 7094 case was technically superior but the decision rested on whether the demands were supported. If Authority use of Atlas was limited to one shift then Culham would probably continue to use Stretch at least for some of their problems. A decision on this need not be taken until the end of the year.
18. Dr. Curtis said there was some obscurity in the costs of machine time on Atlas. Was four-shift working assumed, was the cost of programming included, and so on? The Chairman asked Dr. Howlett and Mr. Hudspith to prepare and issue a note on the cost of using Atlas and what these costs included. Dr. Roberts thought that the whole question of costs needed investigating. Individual establishments wished to use the cheapest machine, yet may be forced onto an expensive machine in the interests of the Authority as a whole: some form of equalisation should be introduced. The Chairman recognised the validity of this statement but thought that this matter could be handled by Group accounts. Action: Dr Howlett and Mr Hudspith
19. The Chairman asked Dr. Marshall to set up a small working part to consider the Authority use of Atlas: requests for time on Atlas, discussion on methods of payment and so on should be channelled through this working party and not directly to Dr. Howlett. Action: Dr Marshall
20. The Secretary asked for a decision on machine allocation for this year. Dr. Marshall said that a decision could not be taken until the 7094 question was settled: if the 7094 conversion was to be carried out then Harwell would start developing more programs on the 7090. After some discussion Dr. Marshall agreed that Harwell could, with some loss of efficiency plan certain programs to run on both Stretch and the 7090. The following allocation was then agreed:-
Harwell 13 hours Winfrith 25 hours Risley 84 hours Culham none TOTAL 122 hours
The Risley allocation covers all the Northern Groups. If the machine is available for less than 122 hours in any week then the allocation is pro-rata: if more than 122 hours are available then any deficit from previous weeks will be made up, pro-rata. If the average available time is greater than 122 hours then Winfrith and Risley should agree on the use of the excess time. This allocation comes into full force when the 7090 goes on three shifts. Until then machine time will be allocated pro-rata.
21. Dr. Corner asked what Stretch time would be required for Civil work. The estimates given were - year ending April, 1964, 24-29 Stretch hours per week, in the following year 23-37, no estimate could be made for the succeeding year.
22. Dr. Howlett reported that Edinburgh University were setting up a data link to the Manchester Atlas, if it was successful they would eventually transfer the line to NIRNS. NIRNS was contributing £3,000 to the total cost of £8,000. Edinburgh would use 1 to 2 hours a week on Atlas. Dr. Howlett had been asked by the Atlas Computer Committee to obtain a technical report on the use of data links based on A.E.A. use. Dr. Corner said that he had been urging for some time that A.E.A. should write about its experience with links. The Chairman agreed with this and asked Mr. Mummery to organise it. The suggestion is that Mr. Barnes of A.E.R.E. together with someone from Winfrith and possibly Risley should write a paper covering all the aspects of our experience. The paper will be sent to members of the Committee for information, but the prime aim is to publish it in a technical journal. Action: Mr Mummery
