Report of Computer Working Party Meeting

17 December 1962

AERE, CPC

Present:

• Mr. Armstrong, AWRE
• Mr. Glennie, AWRE
• Dr. Lance, Winfrith
• Dr. Black, Risley
• Dr. Roberts, Culham
• Dr. Gaunt, London Office
• Mr. Curtis, Harwell
• Dr. Marshall, Harwell

This working party met to consider a number of points in detail and to report to the C.P.C. their recommendations. The questions considered were:

1. What are the consequences of the failure of Fortran IV on Stretch?
2. What time will the Authority need on Atlas for computing?
3. Should the IBM 7090 be converted to a 7094?

It was easy to answer question 1 because in the last few weeks it has become clear that Glennie's S2 compiler will be very efficient and will give Stretch a speed ratio between 2.5 and 3.0 compared to the 7090. The meeting decided for planning purposes to use a factor of 2.5 for 63/64, rising to a factor of 3.0 afterwards. The importance attached to the prompt arrival of S2 was emphasised and we recommend that the C.P.C. formally acknowledge this so that I.B.M.(U.K.) can be persuaded to help in writing S2. Harwell will make available a man to be attached to Aldermaston for six months to help in this work. We have every reason to suppose S2 will arrive on the target date of April, 1963 and, at the latest, June, 1963. Subsequently it will be: gradually improved. In these circumstances the only effect of the failure of Fortran IV is to present Harwell with an unforeseen recoding problem. The tedium. of this recoding will be largely avoided by the use of the 1401 programmes which have been prepared by Winfrith for semi-automatic translation from Fortran II to S1.

Alick Glennie had written a Fortran compiler S1 in response to the late delivery of the IBM Fortran IV compiler. The new compiler was called S2. As a result of this meeting IBM seconded two staff full-time and parts of others to work on S2. These were led by John Nash. Harwell seconded F R A Hopgood to Aldermaston.

The meeting found it impossible to consider questions 2 and 3 in isolation and decided it was advisable to foresee as far ahead as one year after the installation and working of Atlas. It therefore made estimates of the Authority's need for computing time through 63/64, 64/65 and 65/66. These estimates are quite firm for the first two years and the empirical extrapolation made into the third year is reasonable for planning purposes. The following assumptions were used:

1. Stretch will be 2.5 times faster than a 7090 in 63/64 but thereafter S2 will gradually improve so that by April 66 a speed factor 3.0 should be used.
2. Atlas will be 3.0 times faster than a 7090 and therefore Stretch and Atlas time are equivalent. (This last is a simplification because for some work Stretch is still to be preferred because of the larger core store and the peripheral equipment it has).
3. Requests could be divided into three parts: the time on Stretch or Atlas regarded as essential; the time on the 7090 regarded as essential; the time which could be distributed arbitrarily between the computers.
4. That the January C.P.C. meeting will make a firm decision whether or not to convert the 7090. This point is important because although an establishment may declare it has no strong preference between x Stretch hours or 2.5 x 7090 hours it is not practicable to give it x Stretch hours for part of the period and then suddenly replace it by 2.5 x 7090 hours for the remainder of the period.
5. That Atlas will be available for use from April 1965.

The detailed requests follow in hours per week: on the respective machines. Note that in 63/64 the only significant change from the figures considered by the September CPC meeting are an increase in Harwell's request. This increase is required to amend their previous omission of the needs of the Shielding Group (Winfrith staff working at Harwell) and to take account of demands which the divisions concerned had not foreseen four months ago.

		One or Other
Establishment	7090 Essential	7090 Possible	Stretch Possible	Stretch Essential	Atlas Essential	Stretch or Atlas Essential
By April 1963
Harwell	13	5	2	4
Culham	0	0	0	10
Winfrith	25	7	3	4
Risley	84	0	0	0
Aldermaston	0	0	0	1
Total	122	12	5	19
By April 1964
Harwell	13	5	2	7
Culham	0	0	0	10
Winfrith	25	26	10	4
Risley	84	0	0	0
Aldermaston	0	0	0	1
Total	122	31	12	22
By April 1965
Harwell	13	8	3	9
Culham	0	0	0	10
Winfrith	25	42	14	4
Risley	84	0	0	0
Aldermaston	0	0	0	1
Total	122	50	17	22
By April 1966
Harwell	5	0	0	0	30
Culham	0	0	0	0	0	20
Winfrith	25	42	14	4	0	0
Risley	84-100	0	0	0	0	0
Aldermaston	0	0	0	1	0	0
Total	114-130	42	14	5	30	20

Notice from these tables that if we take 122 hours a week as saturation for the 7090 then it is saturated throughout the entire period by work for which it is essential. The earliest date at which the 7090 could be converted to a 7094 is March 1964, and therefore the choice indicated by the central two columns is a real one only if:

1. in 1963/64 the 7090 could be run at an average of 144 hours a week, or, alternatively, that work scheduled for 63/64 could be postponed to the months immediately after the conversion in March 64.
2. in the following years, the 7090 had been converted to a 7094.

Notice also that if the 7090 is not converted the demands on Stretch will be 24 hours per week in April 1963 rising to 34 in April 64 (28 hours per week are known to be available) and rising to 41 in April 65 (40 are expected to be available at that time).

We conclude from this that the two alternatives:

1. to convert the 7090 to a 7094 and discourage the use of Stretch
2. not to convert the 7090 but to use Stretch as much as is required,

are both reasonable plans so far as computing capacity are concerned although plan (i) would give difficulties because the conversion cannot be carried out until March 63 and plan (ii) may give difficulties if it is not possible to correlate the build up of civil work with the availability of Stretch time.

Notice also that the Authority's request for Atlas time builds up very rapidly in 65/66 and by April 1966 Harwell and Culham together request 50 hours a week; we can anticipate this joint request to rise to two shifts (80 hours) two years later. If facilities are provided for using Atlas on line to experimental equipment then time for this is additional to these requests. Because of the commitments the Atlas Laboratory has to other users we feel sure that the maximum time it will be able to allow the Authority is 2 shifts and it is possible that it will have to restrict the Authority to only one shift. We recommend that the C.P.C. request the Atlas Computer Committee for an early ruling on this matter and we also recommend that

1. if only one shift on Atlas is available to the Authority that this be reserved for Harwell and that Culham should continue to use Stretch
2. if two shifts on Atlas will be eventually available to the Authority that both Harwell and Culham should use it and that the Atlas computer committee be invited to consider technical improvements and extensions to Atlas which would make it suitable for Culham's work. These technical matters are discussed in the C.P.C. paper presented by Roberts.
3. Winfrith should not plan to use Atlas because the Authority's share of Atlas time will be rapidly saturated by Harwell alone or by Harwell and Culham together.

The financial considerations concerning the possible conversion of the 7090 are as follows. The conversion would give a saving, on average, of 14.5 Stretch hours per week throughout 64/65; and of 15.5 Stretch hours per week throughout 65/66. From these two years alone therefore a total of 30 × = 1560 Stretch hours can be saved: at the present marginal cost of Stretch this is a saving of £310,000 and hence the capital expenditure of £253,000 for converting the 7090 seems justified. But this is not the entire financial case in favour of conversion because additional savings of Stretch time are possible from the work scheduled for 63/64, and for the years beyond April 66. In the year 63/64 a potential saving of an average of 8.5 Stretch hours per week, i.e. £88,000 exists. However in this year the 7090 will not have been converted and therefore could not take the increased load proposed for it. However financial saving can be made to the extent that the 7090 can be run more than 122 hours per week or that work scheduled for the end of 63/64 can be postponed a few months until after the conversion has been done. By these two means a further saving of possibly £50,000 could be made. A potential saving also exists in the years beyond April 66 because we see no prospect of Winfrith having regular access to Atlas and therefore it is obliged to use a converted 7090 or Stretch. To set against these arguments, the rental for Stretch may possibly be reduced if an extension of the contract is negotiated and in that case the financial case for converting the 7090 is weakened. Nevertheless we consider that the financial case in favour of converting the 7090 is reasonable. In addition to the financial argument in favour of converting the 7090 the Working Party felt very strongly that it was prudent to allow a contingency for unforeseen demands for computer time - previous experience has shown estimates are always underestimates. Indeed, if the conversion of the 7090 is not ordered then we should have in reserve no computing capacity whatsoever until the arrival of Atlas. We should also guard against the possibility that Atlas may arrive late, i.e. after April 1965.

RECOMMENDATIONS

1. That the C.P.C. acknowledge the importance attached to the prompt arrival of S2 and urge I.B.M.(U.K.) to help Mr. Glennie prepare it.
2. That the C.P.C. request the Atlas Computer Committee for an early statement on the amount of Atlas time available to the Authority for computing purposes, and that it inform the Atlas committee that if it were possible the Authority would like a buildup to 2 full shifts after about three years of operation.
3. That the C.P.C. agree that Winfrith should not plan to use Atlas on a regular basis.
4. That the C.P.C. agree that if only one Atlas shift is available to the Authority that Harwell should have prior claim on it and that Culham continue to use Stretch.
5. That the C.P.C. agree that if two Atlas shifts are to be available to the Authority that Harwell and Culham should use it and that the Atlas Computer Committee be invited to consider technical improvements to Atlas which. Culham would find desirable.
6. That the C.P.C. recommend that the Risley 7090 be converted to a 7094 as soon as possible and, in connection with this, that Risley be asked to run the machine as many hours as possible in 63/64 and that furthermore Risley and Winfrith be asked to consider if any part of their work which is scheduled for 63/64 could be postponed to the months immediately following the conversion to a 7094.

W. Marshal, A.E.R.E. Harwell, 2nd January 1963