36. Most established Universities have the basic capability of teaching computer programming in some language or other and the result is that there is emerging la rapidly increasing number of computer programs* requiring testing and further development. This is the point at which the current frustration and dismay in the development of computing work begins; in order to develop a program of moderate complexity it is essential to have access to a computer for short periods several times daily. At present it is not unusual to have to wait a week or more between each short testing run because of the pressure on the computer services, and those Universities equipped with no central computer having to use a remote installation are in an even worse position. One immediate effect is that the problems really needing solution tend not to be attempted; at best much cruder versions are substituted, resulting in weaker results of doubtful value and much dissatisfaction. Even those who persevere and eventually produce a working program, perhaps by spending time and money at a distant installation, now need computer time to run large numbers of cases to test the validity of a theory or to evaluate an experiment. The state of affairs at many Universities is that this production work is postponed for long periods because the computer service is attempting to provide a reasonable turn-round time for short runs. Difficulties arise continually when a piece of computer work is too large or complex for the local installation to handle; the work then has to be run at a larger installation elsewhere, but this usually turns out to be difficult or impossible because the two machines are not compatible. At well over half the Universities and C.A.T's the standard sort of problem requiring quick access time and fast arithmetic (such as matrix inversion and the solution of partial differential equations) cannot be tackled adequately or in a reasonable time.
37. The difficulties of using a computer service are much increased if, as is often the case, especially with university equipment, the computer is unreliable or the maintenance is inadequate. It is almost impossible to perform long production runs when the machine operates without fault for only 85 per cent of its scheduled time as was common enough recently. A much higher efficiency is something that all computer installations should aim for. Indeed, an efficiency of 97 per cent has been achieved by some installations.
* In this report, the convention is used that a programme is a plan of action by which an aim is realised, (e.g. for installing computers in Universities and Research Institutions) while a program is used in the narrower sense of a computer program (e.g. a set of instructions for a computer).
38. Very few Universities now possess a computer adequate in size and reliability for their essential needs. The Universities selected for special technological developments and the majority of the C.A.T's are in an especially difficult position because they are, or should soon become, major users of large and fast computers. But they are not alone; the Universities placing special emphasis on economic studies can also be expected to place very heavy stress on computer facilities of the same kind.
39. The situation is best summarised by saying that the present provisions, and example elsewhere, have created a demand far beyond what can be satisfied at the present time.
40. Many of the above difficulties have been caused or aggravated by the failure to appreciate the very large growth rate in computer usage. Present and projected computer demand for three very large and technologically advanced foreign laboratories are shown in figure 1. After some initial hesitation they appear to settle down to growth rate factors of 2 per annum, or even higher. The growth in demand for computer time at a number of the larger British Universities has been analysed. Allowing for sharp changes at the introduction of new facilities, the growth rates are commonly found to lie between a factor 1.5 and 2.0 per annum, equivalent to a doubling time of between one and two years. Such rates of rise may seem surprising, but there is no indication of any slackening off as yet. Although in the past the physical and engineering sciences have no doubt provided the driving force, our survey of present and projected usage has made it obvious that economic studies, medical applications, and information retrieval of all kinds will maintain the rate for a considerable time to come.
41. This heavy rate of growth must always be borne in mind in planning the operation and equipping of general computer services at Universities and elsewhere.
42. A doubling in demand does not, however, imply a doubling in cost because of marked economies in the design of newer and larger computers. In order to determine the effect of technological advances upon the effectiveness and cost of computers an analysis was recently made in the United States of four successive computer models offered by the same manufacturer over a period of seven years*. The results are given below, assigning the first computer (A) a base figure of one:
| Computer | Effectiveness | Rental Cost | Effectiveness/Cost |
|---|---|---|---|
| A | 1.00 | 1.00 | 1.00 |
| B | 2.52 | 1.86 | 1.35 |
| C | 8.16 | 2.57 | 3.18 |
| D | 12.41 | 2.79 | 4.45 |
Thus the effectiveness of these computers was increased by more than 12 although the cost rose less than three times. An annual doubling in demand can thus be met by a much lower financial growth rate.
* Report to the President on the Management of Automatic Data Processing in the Federal Government, issued February 1965 by the U.S. Bureau of the Budget.
43. Recognition of the growth in demand has led the computer industry recently to introduce modular machines with standard interface arrangements so that the installation can be continually upgraded within a very wide range to meet rising demand and new requirements.
FIGURE 1. Growth of Computer Use at High Energy Laboratories.
(Extract from "Computing Needs for Argonne for Fiscal Years 1965-68"
by W. F. Miller.)
44. It must be admitted, however, that Universities have not uniformly made the best use of the limited resources available to them, nor has the industry done as much as it could to provide the ancillary items essential to a good service. The chief lack is of course in sheer computing power, the main burden of this Report; in the remainder of this chapter, however, we dwell on such matters as the need for better peripheral equipment, the problems of compatibility and software, the staffing and status of a computer service unit within the university, building requirements, and some desirable features which an adequate service might be expected to have in the future.
45. Problems requiring large amounts of input and output data require that good card punching, duplication and verification facilities be available; these are provided at only a few installations, and even at these large quantities of essential input data have to be punched by academic staff who would be much better employed otherwise. Associated with large input is large output; even though the central processing unit may be capable of handling the required calculation, the absence of an adequate line printer and card punch very often prohibits proper use of the whole facility. This is the kind of work needed to be done by the teaching hospitals, for example, in the application of information retrieval to medicine, pharmacology and surgery.
46. It is not sufficiently understood in the Universities that different problems require different treatment and thus differing peripherals. Certain problems are most conveniently handled in terms of paper tape; others fit more naturally into a card-oriented facility. A general computing service to be useful and efficient must be able to handle both with ease.
47. Adequate and varied peripheral equipment takes up almost as much space as the computer itself. Far too often the space allotted has been underestimated, and the confusion which results is sometimes beyond belief. The user's contact with the computer is solely through the peripheral equipment; inadequacy, malfunctioning and poor organisation of peripherals are thus his most obvious source of frustration.
48. When a programmer is told that his own installation is inadequate for one of a number of reasons he often finds that a larger machine elsewhere while big enough for his purpose cannot be used either, without excessive labour, because the language he has used on his local machine is not provided for on the larger one and rewriting a major program can take weeks or months. This occurs frequently for example at Universities and C.A.T's possessing Elliott 803 machines but wanting to use the Atlas, IBM 7090 or KDF 9 machines. There are also many programs already written in Fortran II and IV and tested in the U.S.A. which could be made available with advantage in this country but which cannot be used because some of our machines have inadequate Fortran facilities.
49. Lack of appreciation of the importance of efficient software on the part of industry and user alike has led to a proliferation of programming languages, and dialects of languages, and has made the running of quite simple programs on different machines, or even the same machine with slightly different configurations, impossible. The significant scientific languages at the present time are Fortran, Algol and, to a lesser extent, Lisp. It is possible that the new IBM language PL/l (originally called NPL) will become widely used in the future. On the Chilton Atlas compilers are provided for about a dozen languages. Writing compilers for some of these languages is wasteful and merely encourages the use of inefficient programs. We have investigated existing compilers in detail and have concluded that it would be better to have two or three really good compilers than twelve poor ones. On the KDF 9, two compilers exist for Algol, one for a form of Mercury Autocode and one at the instigation of the Atomic Energy Authority, for Fortran. The Algol compilers for KDF 9 are less than satisfactory. One produces code that is only 2 per cent efficient; the other is moderately efficient (30-40 per cent) but it takes several minutes to compile even very short programs. The Fortran compiler and Egdon operating system recently written for the upgraded KDF 9 machines of the A.E.A. are quite satisfactory, and enable this machine to operate with a power comparable to that of an IBM 7090 machine or better; the existing Algol compilers reduce the power of a university KDF 9 to something comparable only with Mercury.
50. The Egdon system and Fortran compiler recently provided for the A.E.A. machines will not fit into the smaller machines at the Universities, although they will do so if these machines are upgraded as recommended later in this Report. In view of the emphasis placed upon Algol at many universities, however, it is important that a much improved Algol compiler should also be provided for these machines. Furthermore, it should have some form of card implementation as well as the usual 8-hole paper tape form. An Algol compiler satisfying these criteria has been developed at Munich for the IBM 7090 and is said to be as efficient as Fortran in both compiling and execution speed.
51. We accept the fact that one of the principal tasks of computer research is to develop improved languages which will gradually improve the efficiency with which computers are used. Nevertheless, the common use of a variety of languages and dialects is to be deplored since it increases the problems of incompatibility. The solution lies in ensuring that every computer service unit is provided with, and maintains, the few key compilers that are accepted at anyone time in addition to experimenting with new ones under development.
52. Software is likely to become more critical than hardware once the new equipment now on the market has been assimilated. We therefore attempt to lay down what we consider to be the minimum software requirements for machines now being developed or to be developed in the near future. Any scientific software package being defined at the present time should include the following:
53. Experience with computers so far has shown that the use of magnetic tapes for common input/output of many programs causes inefficiencies which might best be eliminated by the use of a disc. A priority system of scheduling should be provided by which short jobs can be given preference, thus speeding up program development. Since compilation of individual subroutines can now take as little as a few seconds it is not unreasonable to expect a turn-round in the development stage of a few minutes. The organisation phase (which includes calling all appropriate subroutines from the disc) at the beginning of each job should be as efficient as possible, and this is particularly significant for short jobs. Time sharing facilities should be exploited so that the idle time of the central processor may be kept to a minimum. Facilities should be provided by which any material, e.g. routine libraries, can be stored on the disc and called into use automatically.
54. Since even a moderate computing system can be so complicated it is extremely important to have good diagnostic facilities for programmers, operators and engineers.
55. We have called attention to the need for compatibility in the matter of language. Compatibility between different machines is also greatly facilitated by having certain standard items such as punched cards, paper tapes, magnetic tapes and discs which can be transferred from one computer to another, thus saving much additional time and effort. At present such items are not sufficiently standard.
56. It is possible using modern languages to do very complicated jobs in an economical amount of machine time: it is, however, only too easy to use large amounts of machine time to do quite simple jobs. Experts are armed with a knowledge of what the machine compiler facility will make of a particular piece of coding and avoid using phrases and passages which the compiler would translate inefficiently. More important perhaps is that the expert will avoid using ungainly methods and faulty organisation in programs which may eventually produce the correct results at the expense of using arithmetic units, tape or disc units and peripheral units in a grossly inefficient manner. The teaching role of the service is thus an important one if machine time is not to be wasted.
57. There are many points between the preparation of a computer program in a particular language and a set of correctly computed results at which errors can arise. The punching of programs and data for example is essentially a job much better done by people who are paid to do a lot of it on a permanent basis. A trained punch card operator paid a normal clerical staff salary can produce a perfect deck of cards, or roll of tape, in a fraction of the time an academic would take to produce a deck or tape, full of errors, which he then has to correct, more often than not in a primitive manner. Full time operators can purge programs of syntactic errors.
58. In the larger central computer installations there has to be a well defined organisation for handling:
59. For each central installation this work is best planned initially by senior computer staff and then carried out by trained operating staff who run the system according to well formulated and established rules. Again, experience shows that if the planning is good, quite junior operators can run very complicated and difficult programs with no trouble.
60. We have paid especial attention to the problems of the administration of the computer service within a University. Practice differs quite widely at the present time: some Universities have organised their computer service within a Department of Electrical Engineering or a Department of Mathematics, some within the Faculty of Science, some under the loose control of a professorial committee. The director of the service sometimes is a professor in his own right, or has equivalent status, sometimes he has a much lower standing in the University.
61. A properly equipped computer laboratory, adequately staffed, is comparable in size to many science departments, and more expensive than some. The cost of a medium sized computer is comparable with, for example, a small nuclear accelerator or a radio-telescope. Moreover, it is not desirable that the operations of a computer service unit should be too closely dominated by the thinking of any single department. To give only two examples of attitudes which have hindered the development of a computer service, some departments of engineering tend to be too interested in the hardware aspect to give proper attention to software, whereas some mathematics departments quite properly feel that computing is to be regarded by them as the last resort.
62. We are convinced that a computer service unit should be a department in its own right, with its own staff and budget, and that its director should have a seat on the appropriate committees of the University, including Senate, much as does the University librarian. Like the librarian he should be advised by an inter-faculty committee to ensure the proper development of computing as an interdisciplinary activity.
63. Running a computing installation of the power of (say) a KDF 9 or ICT 1907 is a major undertaking. To keep such a machine busy on two or three shifts for fifty-two weeks in the year, providing a service to customers requires a large number of good quality staff at a wide variety of levels. To do less is to waste the machine. Finding these staff might well prove the biggest difficulty in setting up new computing laboratories or expanding existing ones. Computing laboratories may require staff in any of the following categories: director of laboratory, programmers, main frame operators, punch room operators, machine and punch room reception and dispatch, teaching staff, general services, and maintenance engineers.
In some University laboratories where there is a separate Professor of Computing Science (or numerical analysis, etc.) the teaching functions and perhaps some programming may instead be placed under his control.
64. Even the smallest laboratory will require a Director and main frame operators with an engineer on call. The larger University Computing Laboratories will probably find, as the Government Laboratories found several years ago, that they require staff in all eight categories, although the maintenance engineers are nowadays usually provided by the Company supplying the computer.
65. Upon the personality and competence of the Director the whole success of the laboratory will depend. For all but the smallest installation he ought to be in the Reader or Professor salary range although academic distinction need not be his chief attribute: it is most important that he should be a good manager of the service. He ought to have been appointed to his post as early as possible in the period of build-up of the laboratory: preferably before decisions as to what computer, size of staff, buildings etc., are made. He should have had several years really practical experience of computers and should regard the training of programmers as an important duty. If he has a line of research of his own which he wants to follow up on the machine so much the better. Opportunity to do some research as well as running a service will help attract a good man. Understanding of customers' problems is greatly helped by keeping in close touch with the nature of the problems as well as with current developments in the installation. He will find that tact is one of his greatest assets.
66. Such people are hard to find but no choice is more important for a computer installation. A first class Director will not only build up the laboratory sensibly but he will also attract good staff to back him up.
67. The minimum number required in this category is zero for a small laboratory run on an entirely" open-shop" principle where users normally prepare their own programs. In practice it is almost certain that, even in the smallest laboratory there will be one or two programmers available for assistance and advice; for example the Director of a very small laboratory might fill the dual role of Director and Senior Programmer. If the laboratory has a powerful machine, or offers wide service, or runs on mainly closed shop lines, then the programming group must be made larger and more professional. There is no maximum to the size of the group. An installation run on entirely "closed shop" programming lines, having a wide research programme of its own and a machine of the power of a CDC 6600, could easily require 50 programmers.
68. General experience is that an installation suits its advisory service, programming assistance and research effort to the number of programmers available. There is a great shortage of high quality programmers and this shortage is liable to become worse as more and larger computing laboratories are set up. It is recognised that a good programmer can effectively earn his keep several times over, his more efficient coding of programs saving tens or even hundreds of hours of computer time each year.
69. The number required will clearly vary with the size of machine, nature of the work-load and number of shifts worked. There is a general tendency to run the short jobs during the day, including most of the development runs. Thus the crew required on the main frame during the day shift may be more numerous than on the evening or night shift. A rough formula indicating the number of operators required on the day shift of anything other than a very small machine is as follows:
Let: a = number of input peripherals (card readers + paper tape readers),
b = number of magnetic tape decks,
c = number of printers,
d = number of card punches + paper tape punches.
Then the number required on the day shift is approximately:
N = 3/4 a + 1/6 b + 1/2 c + 1/3 d + 1
The last term is to allow for a Shift Supervisor.
For example, in the case of the Chilton Atlas Laboratory
a = 4, b = 18, c = 2, d = 3, and so N = 9.
In fact the Chilton Atlas Laboratory has 8 operators on each shift.
Again, for a KDF 9 with 8 tape decks, 1 printer, 1 card reader, 2 paper tape readers, 1 paper tape punch and 1 card punch.
a = 3, b = 8, c = 1, d = 2, and so N = 6.
70. If the policy of the laboratory is that everyone must punch his own program then sufficient auxiliary equipment (card punches, flexowriters, etc.) must be provided to minimise delays. Few laboratories will take such an extreme view and even a small machine (such as an ICT 1904) ought to be supported by two or three punch room operators. For a large installation (Atlas, CDC 6400) which offers a punching service to customers, 12 or 15 punch room operators might well be required. In any case, a few card punches and flexowriters should be provided for the use of customers who have only a small amount of punching to do. The operators can then be used for the bigger jobs.
71. If the computing laboratory provides a service to other departments or Universities they will quickly discover that the handling of incoming and outgoing cards, tapes and documents is a full time job for at least one person, and perhaps two. These reception/dispatch assistants will need to keep very clear records, logging in jobs as they are received and ensuring that the jobs are actually performed and the input and output is returned complete to the customer. It is surprising how many programs are mutilated or lost within the average University service unit due to lack of attention to such apparently trivial matters.
72. Almost all University computing laboratories, even the smallest, will be involved in teaching. It is quite impossible to lay down any general rules about this: there may be Courses on programming, numerical analysis, computer applications etc. and these may be the responsibility of a Professor of Computing Science (or numerical analysis, etc.) specially appointed for this purpose, or of the Mathematics Department, or the Director of the computing laboratory.
73. If the computer is to be efficiently used it is essential that proper courses are given to familiarise users at all levels not only with programming methods in general but with the facilities of the laboratory itself and with arrangements with other laboratories. The more senior members of the laboratory should, therefore, wherever possible, become members of the teaching staff of the University.
74. The size of this group is governed by the total number employed in the other groups and by the kind of service the laboratory is offering. If it expects a lot of visitors to come and use the machine and offers to find hotel accommodation, provide transport to and from railway stations, etc., then several full time staff will be required. Other types of staff coming under this heading are secretaries and typists, cleaners and watchmen. A minimum requirement in this class is one person (secretary to the Director), but a large laboratory having hundreds of users may require a dozen people for general services.
75. By way of illustration of some of the ideas put forward above consider two installations. These are reasonably typical of (1) a normal University installation; (2) a large installation providing a wide service and having some research lines of its own.
Case 1. KDF 9 having 3 tape decks, 2 readers, 2 punches,
1 line-printer.
The formula for main frame operators suggests about 4.
Thus, for 1 shift, the staff and salaries might be:
Director 1 × £3,500 £3,500
Teaching/Programmers 4 × 1,500 6,000
Main frame operators 4 × 750 3,000
Punch room operators 3 × 500 1,500
Reception and Dispatch 2 × 500 1,000
Admin. (inc. Secretary) 2 × 800 1,600
Total 16 £16,600
By comparison, Oxford which has exactly the KDF 9 specified above, has a total of 14 staff, viz:
Director 1
Teaching/Programmers 5
Main frame operators 3
Punch room operators 3
Research Assistant 1
Secretary 1
Total 14
For an upgraded KDF 9 on 3 shifts another 10-12 staff will be required. The salaries bill would then be about £30,000 per annum.
Case 2. Atlas Laboratory, S.R.C. The machine has 18 tape decks, 4 input readers, 3 output punches and 2 line printers. As shown above this indicates 8 operators required for the main frame. The total number of staff required to run this machine on 2 shifts is:
Director 1
Programmers 10
Main frame operators 16
Punch room operators 9
Operations, Programming, Records 4
Research Fellows 4
Administration 10
Total 54
The total salaries bill is about £70,000 per annum.
76. Thus, computing laboratories require staff of many different kinds ranging from a Director to machine assistants and cleaners. An upgraded KDF 9 will almost certainly require a total staff of 25 or more. The salaries of these people involve a recurrent sum in excess of £30,000 per annum.
77. We have already touched upon the inadequacy of the space provided by many Universities to house the computer, its ancillary equipment and its staff. Installations vary very greatly in size, configuration and complexity; moreover, all these attributes change with time, especially now that modular series of machines are being introduced. In the future no installation is likely to retain its initial configuration unchanged for more than 2 or 3 years, as will be seen from the recommendations made later in this Report. This means that buildings should be planned with great care and with an eye to flexibility and future expansion.
78. Expert advice on the accommodation and installation of computers can be provided by the Computer Advisory Service of the Ministry of Technology. The rough guide-lines which follow are based on their experience.
79. In table 1 we show the size of room required to house a range of typical computers, together with power requirements, and air conditioning requirements. To the total areas represented in the table should be added at least an equal area for input data preparation, receipt and dispatch and associated clerical work, and at least an equal area for office accommodation, stores, welfare services, and corridors.
| List of Machines |
Power Requirements (Note 4) |
Heat Dissipation B.T.U./hour |
Air Conditioning Requirements | Size of Computer Room |
Maintenance and Air Conditioning Plant Room |
||
|---|---|---|---|---|---|---|---|
| Temperature of |
Relative Humidity |
Filtration % at N microns |
|||||
| Very Small | An engineers maintenance room [say 200-500 ft2] and an air conditioning plant room [say 400-1,500 ft2] will be required. The size of these rooms will be roughly proportional to the size of the computer configuration. | ||||||
| Elliott 803 | 10-15 KVA | 27,000- 41,000 |
70±3 | Note 2 | 90% at 5 | 600- 800 sq.ft. |
|
| Elliott 4120 | 45%±5 | 95% at 5 | |||||
| ICT 1902 | 50%±10 | 95% at 5 | |||||
| ICT 1903 | 50%±10 | 95% at 5 | |||||
| ICT 1904 | 50%±10 | 95% at 5 | |||||
| IBM1401 | 70±5 | 50%±10 | 90% at 5 | ||||
| Small | |||||||
| Elliott 503 | 15-18 KVA | 41,000- 49,000 |
68±3 | 55%±5 | 95% at 5 | 1,200- 1,500 sq.ft. |
|
| ICT 1905/1 | 70±3 | 50%±10 | 95% at 5 | ||||
| IBM 360/30 | 70±5 | 50%±10 | 90% at 3 | ||||
| IBM 360/40 | 70±5 | 50%±10 | 90% at 3 | ||||
| Medium | |||||||
| KDF 9 | 25-30 KVA | 66,000- 82,000 |
70±3 | 50%-60% | (99% at 5) (90% at 1) |
1,800- 2,000 sq.ft. |
|
| ELLIOTT 4130 | 70±3 | 40%-50% | 95% at 5 | ||||
| IBM 360/50 | 70±5 | 50%±10 | 90% at 3 | ||||
| ICT 1907 | 70±3 | 50%±10 | 95% at 1 | ||||
| Large | |||||||
| CDC 6800 | 40-70 KVA | 109,000- 191,000 |
?(Note 3) | ? | ? | 2.000- 3,000 sq.ft. |
|
| IBM 360/92 | 90±5 | 50%±10 | 90% at 3 | ||||
Note 1. In all cases the power and space requirements will depend on the number and type of peripherals installed. there are also slight variations between different manufacturers for the same class of machine.
Note 2. Humidity control not required because no magnetic tape units available; only magnetic film handlers.
Note 3. Air conditioning requirements unknown but will probably be similar to those for the IBM 360/92.
Note 4. Usually 240V A.C. ±10% 3-Phase 50 c/s±1%. Sometimes a motor alternator set may be required, dependent on the quality of the mains at the particular site. The C.D.C. machine may in any case require a m.a. set to convert from 50c/s to 60 c/s.
TABLE I: ACCOMMODATION REQUIREMENTS FOR A RANGE OF TYPICAL COMPUTERS
80. A very rough estimate is that the building costs are about one-sixth of the capital cost of the equipment.
81. We have already stressed that the development of a computer program requires frequent testing, either in whole or part, and frequent modifications. Sometimes after a testing session, it is clearly obvious that an elementary and easily corrected mistake has prevented the whole program from being operated by the computer. The programmer is now anxious to deal with the correction (sometimes by punching one card only, a job completed in seconds), and to get back to running it again on the computer. At this stage, the pace of development is critically dependent on the " turn-around" time, that is the time between the points at which the programmer's work is submitted to the computer centre, and the return to him of the data on the completed test. At a few centres in Britain it is possible to do three such tests per day; at many others, and in Universities generally, the turn-around time is much longer and in some cases a week is not unusual.
82. The Americans recognised this need for quick testing and have conceived at Massachusetts Institute of Technology, an experiment for extremely rapid access and turn-around time. It was conceived further that programmers may be remotely situated and need quick service by telephone link. The MAC (Multi-access Console) project at M.I.T. consists of a large central computer, in fact an IBM 7094 Central Processing Unit with 64K of magnetic core storage, disc files of 4.6 million words, plus central punch, printer, scope, and magnetic tape facility. By MAC type channels it is planned to serve up to 60 on-line typewriters. A complex software system consisting of supervisors and compilers is being developed to run the central computer much like a central city telephone exchange. This software is very complicated because a large number of simultaneous users must be envisaged and catered for in the system. The software must be in control of priority and time-sharing; the compilers must allow the use of a variety of languages.
83. Briefly, the user carries out the following procedure:
The system is particularly good for small programs or short sections of large programs with short running times (5 seconds or less) with short outputs (100 seconds or less). The system is subject to a number of objections, a serious one being that sooner or later a programmer begins to demand substantial amounts of output and substantial running directions as direct input. The remote typewriters cannot deal with other than tiny amounts of data.
84. This sort of system is however clearly adaptable for more substantial remote users. When it is necessary for such a user to have card input and line printer facilities the remote terminal must clearly be better than a MAC type terminal and a small sate11ite computer can be envisaged as taking its place. Present generation American equipment has been developed, to enable a small machine to " hold conversations" with a big central machine. Further use of the satellite machine is now possible. For example MAC type remote terminals can be connected to disc storage on the satellite machine and the satellite then acts as a filtering device for work which it transfers automatically to the big machine if beyond its own capacity.
85. Hybrid schemes are presently being discussed in this country; one such is a large central facility taking work from MAC type terminals on a University campus together with work from a few satellite machines in the surround, a teaching hospital situated a mile or two from the central facility or a remote College or neighbouring research laboratory.
86. The type of conversation arrangement between the large central machine and satellite machines or teletype terminals will of Course be determined by the arrangement of computing requirements in the area. These needs will change with time and the system must not be rigidly designed. The software work associated with all these systems must not be underestimated; the provision of a supervisor for anyone of them represents a major programming task (estimated for example at 30 man-years, compared to 20 man-years for the Egdon system). There is little doubt, however, that adoption of multiple-access systems in the future will revolutionise the nature of a central computer service.