We consider that the problems currently besetting the provision and organization of computing resources within SERC arise from:
The effect of these difficulties is increased by recurrent financial pressures leading to unplanned cuts in provision, and the net result has been an effective breakdown in planning procedures. We consider that it is imperative to devise and adopt a set of funding and administrative arrangements which are stable, and which will not collapse under minor perturbations. Before suggesting such arrangements, we have considered in detail each of the specific problems referred to above and describe these in the following sections.
We consider that the administrative and funding problems within SERC itself essentially arise from the conflicting demands of Board provision and central support, and they are increased by the splitting of central computing support between establishments.
We consider that the present administrative structure of a Central Computing Committee, a set of Board Computing Committees and a part-time Computing Coordinator is unsatisfactory because:
Two different mechanisms for funding central computing have been tried and both have proved unsatisfactory. Until 1982 the funds required for central computing were provided directly from Council. The level of support initially established was then reduced in order to fund more science and technology. In the absence of a direct voice to defend computing needs on Council it proved difficult to establish a correct level of support. These difficulties were recognised by the Computing Review Working Party which reported in December 1980 and a different method of funding proposed in which computing costs were attributed to Boards. This recommendation was accepted by Council, and Boards have been directly responsible for funding computing since 1982. In this method Boards can vary their contributions over the medium term. This method in turn has proved very unsatisfactory because short-term financial pressures have led to precipitate and uncoordinated reductions in the levels of support which Boards have wished to make. The scale of the resulting divergence from an initially agreed level is illustrated by the current proposals to reduce central support from £10M to £6M over a period of three years.
Furthermore Boards are not willing to pay for the necessary infrastructure costs in any agreed way, and are not willing to make adequate provision for the routine replacement and updating of equipment. There is, in addition, no way of planning and coordinating these large expenditures on major new equipment with other interested bodies.
In the past, these deficiencies in funding provision have been masked by the regular availability of unspent capital, which enabled computer provision of various sorts for central support to be treated as a balancing item towards the end of the financial year, so that regular provision could be made on an ad-hoc, but nevertheless reasonably assured, basis. This convenient cushion has now completely disappeared, with the unfortunate result that an urgent need for carefully planned provision has coincided with a degree of financial stringency which has rendered such an exercise virtually impossible. It is no exaggeration to say that the existing funding mechanisms have broken down completely.
Strategic planning is an important element in optimising the cost-effectiveness and performance of the central computing provision especially in the face of rapid technological change and changing user requirements. Strategic planning is also vital where more than one body is involved in providing computing resources for university research.
Such planning is, however, impossible in a situation where in one year Boards can make precipitate reductions in their provisions for central computing. An effective removal of capital funding means that no provision can be made to replace equipment in order to reduce running and maintenance costs, or to make substantial future provision for the next generation of super-computers. The costs of such a machine, say of the order of £10M at current values, are such that it is unreasonable to seek to solve this particular aspect of central provision within SERC alone. It is necessary to set up a wider framework of provision and collaboration, and this cannot be done without the close involvement of the Computer Board and the other Research Councils, and hence cannot be done without a careful examination of the role of the dual support system in providing computing resources for research. If this particular problem of super-computer provision could be satisfactorily resolved, then clearly it would have a greatly beneficial effect by ameliorating the severity of the problem of working out appropriate funding and administrative arrangements within SERC itself.
As has already been implied, the essence of the problems associated with the split of provision between Boards and central support lies in:
The scientific and technological justification for using more computing power is that it enables research workers to do work which was previously not possible. However the benefits from rapid technical advances in computing are not without accompanying problems.
Apart from the obvious ones associated with rapid obsolescence and the difficulty of making effective long-term plans, there are some less obvious but equally difficult problems. These arise from the every-increasing trend towards distributed computing. In particular the increasing power and cheapness of hardware means that what was natural in central provision yesterday becomes natural in local provision tomorrow. On the one hand this should lead Boards increasingly to make their own provision; on the other hand however it rapidly opens up the possibility of an excessive duplication of effort, particularly in software provision, and of an excessive proliferation of unstandardized hardware. It therefore adds to the conflict between Board autonomy and central coordination.
Some of the problems associated with hardware provision arise from the increasing costs of some types of equipment, such as super-computers, and others, paradoxically it may seem, arise from the rapidly decreasing costs of other types of equipment, such as personal computers and small workstations. Although the cost/performance ratio is improving at the lower end of the range it is this ever-increasing range of costs which is putting a strain on the existing funding and administrative arrangements. Furthermore one cannot expect overall hardware costs to shrink greatly because the volume requirement is growing rapidly, and also the needs of users are requiring more sophisticated equipment.
Although at one time the cheapest form of computing provision for large numbers of users was associated with large mainframes, this is no longer true. However although minicomputers are relatively cheap to buy they cost a great deal in support and in software.
The effects of simultaneously increasing costs for one type of equipment and decreasing costs for other types are combining to force the pace in a change towards distributed computing over networks because:
Hence the principal effect of hardware developments is to require the provision and maintenance of an effective infrastructure in the form of a network together with the skills to run it and develop it. In turn, the emergence of widely-adopted networking standards will change the way in which the various research communities within SERC work, and hence change the organization required to make appropriate provision for them.
The problems associated with software provision essentially arise from:
It has been estimated that it takes one hour for a skilled programmer working on a large and complex program to produce one line of code; many scientific programmes contain 10,000 lines of code.
These difficulties pose severe problems in organizing the effective provision of computing resources across the widely varied activities of SERC's various Boards.
The proper support of users, especially the sporadic user, or the continual user working near the limits of available systems, is rapidly becoming the most expensive aspect of the proper exploitation of a computing environment. The availability of ever cheaper hardware means that the entry-cost for a potential user continues to fall, allowing users to acquire suitable (or unsuitable!) hardware with the minimum of consultation or enquiry. Users often decline to take advice, even though it is free. This forces still higher the unit cost of providing user support, which necessitates having a wide range of expertise to deploy. At the same time at least some users are not making the best use of the skills on offer, and of their own equipment.
The correct balance between encouraging users to experiment, and a judicious resort to expert advice is never easy to achieve. This is especially true in computing, where the technology is thought of as being of interest in its own right and spills over into many areas of instrumentation.
It is often argued that for the sporadic user, provision of proper documentation reduces the need for user support. This conveniently overlooks the problem of who writes the documentation, and of its cost. A page of A4 documentation needs about one man-week to prepare - the same amount of advice in a help file probably needs four man-weeks.
The recent acute short-term financial pressures experienced by SERC have greatly exacerbated many of these problems and have vividly high-lighted the urgent need for their solution. Indeed, as has already been said, the severity of these pressures has led to an effective breakdown in the Central Computing Committee's planning procedures. In that this has forced the radical examination of needs undertaken by this working Party, it is not altogether a bad thing. What is essential, however, is that any new administrative and funding arrangements which are adopted must be sufficiently robust to withstand these sorts of pressure without breaking down.