In considering the problems besetting SERC computing and the various options for their solution we have come to the following general conclusions:
We recommend that Option I, as detailed in the previous section, should be rejected and that Option II should be adopted as a solution to the short term problems within SERC. The longer term problems of providing access to a state-of-the-art computer could be satisfied by either Option III or IV but we recommend that, in view of the costs of such a computer and the possible benefits to the wider Research Council community, Option IV should be accepted. We set down below some detailed proposals concerning the two recommended options.
Option II would, in our view, provide a clearer division of responsibilities between Board/central provision for computing than at present, and would provide a much improved funding mechanism for those parts of the central computing resource which would be provided centrally or shared between Boards. Within Option II, we make the following specific recommendations:
The AS 7000 should be handed over to the Science Board who should operate it as a facility for the on-site analysis of data taken on the SRS and grant related work within the Science Board area if the capacity is available. The funds for computing for the NSF should be made part of the direct resource for that project and those concerned should be free to bid for computing facilities on this machine, and should make direct payment for it. Daresbury Laboratory should control the charging procedure to the projects involved, but the controllers for the relevant projects should be free to decide what level of computing they wish to pay for and use. The Science Board should underwrite the operating costs and should determine the level of manpower provided together with the future capital and recurrent provision.
This service should be run on a direct charging basis. The charging policy used should be aimed at making a sufficient profit to provide for future capital replacement. Council should underwrite the procedure and should accumulate any profit made, seeking approval for greater carry-over powers to allow this to be done. Future capital purchases could be taken from the Council reserve in the relevant years as required. An alternative would be to consider leasing such machines. Should there be an accumulated loss, then these RAL-based systems will have to be reduced until the resources provided match the project requirements needed by the Boards.
If this proposal were to be adopted, then project leaders would be asked in January to bid for their computing requirements for the coming year. They would have to pay for their bid, but the payment would be at a discount rate. Projects could use more time than their bid allowed, but this extra time would be charged at the full rate. Such a procedure should give a fairly specific advanced warning of the likely demands for one year ahead. FYFL bids by Boards would provide some additional guidance. A differential charging rate for different times of use would be introduced to spread the computing load. The finance needed to support this computing would be distributed to Boards and through them to the projects. Administrative computing would be directly charged to the relevant budgets.
The Computing Coordinator should be asked to provide a set of detailed proposals setting out an appropriate charging policy and its means of implementation.
We recommend that the ICF be directly funded by the Engineering Board. Other Boards should be free to bid for time on a repayment basis. The responsibility for support of non-ICF and VAX workstations should be devolved to the appropriate Board, although some mechanisms will have to be found for supporting those workstations which are used equally by all Boards.
Responsibility for support and management of the single user systems programme should become the responsibility of the Engineering Board as the principal user, but Boards should contribute to the costs in proportion to the value of the systems currently installed in their area. The rate of these contributions has yet to be determined but an initial preliminary estimate suggests that the rate might be around 0.25 times the value of the systems currently installed.
We recommend that the provision for a general computing infrastructure and networking should be a direct responsibility of Council.
If the above proposals for funding are adopted, all but the infrastructure support costs will have been returned to the Boards, and there will thus be a direct coupling between the science and technology supported and the computing deemed necessary in order to carry projects out. Only RAL will be involved in computing for all Boards.
We recommend that a full-time Director of Computing be appointed with the following responsibilities:
We recommend that a small advisory body, comprising a number of outside experts and Board representatives involved in a range of research-related computing activities and in computer use and development, should be formed to assist the Director and Council in formulating and executing computing policy. The duties of this advisory body would be:
We recommend that SERC should not in future pay vector processing charges for computing done at the existing National Centres, for the reasons given in Section 6.
We have not had time to work fully through the details of the specific recommendations included in Option II and in particular the specific details of the charging policy which would apply for the RAL mainframes. We recommend therefore that the principle of Option II should be accepted and urgent consideration be given to developing a workable solution taking account of the difficulties and constraints that might apply.
Clearly these new funding arrangements, if approved, cannot be introduced before the beginning of next financial year.
We strongly recommend that every effort should be made to introduce them in the financial year beginning in April 1985.
We consider the use of a computer bureau to provide access to state-of-the-art computing not to be practical and not to satisfy the requirements of users.
We recommend that in view of the demand for large state-of-the-art computers, as identified in other sections of this report and described in Appendix C, urgent consideration should be given to the total national requirement for Research Computing which would include provision for such a state-of-the-art computer. In view of the high cost of such a machine and the heavy burden it would place on SERC if this were to be run for SERC work alone, we recommend that Option III should be rejected in favour of Option IV. Section 6.3 of this report gives further justification for this recommendation. We have not had time to explore this option fully and therefore further recommend that a joint Computer Board/Research Councils Working Party be set up to explore:
