In considering possible options, the Working Party has been influenced by the need to provide a robust solution to SERC's computing problems which will also provide an adequate mechanism for strategic planning, which will cater for the need for computing provision to be project-led, and which will provide a direct and responsive coupling to science and engineering needs. Furthermore the working party was aware that its recommendations, if accepted, would have to be implemented over two very different time scales: the short time appropriate to a solution of SERC's internal administration and funding problems, and the much longer time appropriate to a solution of the strategic planning and super-computer provision problem.
Given that the Working Party's primary task was to address these problems it did not have the opportunity to examine in detail the relationship between SERC and the UK computer industry. It believes however that the relationship does require further examination and in the meantime SERC should continue to support UK industry as it has done in the past.
Areas where SERC may have to provide computing support include:
In the course of our investigations, in broad terms, four general options emerged, which were discussed with users and, in a brief and exploratory way, with external bodies:
This would give the various Boards complete autonomy in providing those computing resources which they see as necessary to pursue their individual programmes of research. Carried to a logical conclusion it would involve SERC only in providing support for central experimental facilities, and consequently no infrastructure or network would be available unless the Boards worked in unison to provide them. In the absence of any substantial central provision, some current grant holders would have to seek support for their computing from universities or National Centres.
Under this option Boards would be made responsible for providing those computing resources which are directly required for their experimental facilities, while some central provision would continue to be made for those aspects of research computing judged to be more cost-effectively handled in this way. There would be central support for the infrastructure and networking but all other computing provision would depend on the Boards.
Here there would be a very substantial central provision, including state-of-the-art computing facilities, although as much facility-related computing as possible would still be devolved to Boards. The very considerable accumulated resource and skill in terms of trained manpower in all forms of scientific computing available within SERC would be a significant national asset, and it would be important that the best possible use was made of it. Consideration should be given to the provision by SERC of a national resource for research computing under a peer review procedure and integrated with existing National Centres. The form of this integration will need careful examination but one solution may be the creation of a third National Centre dedicated to the needs of SERC grantholders. This would give a better working relationship between SERC's central provision and the other national arrangements for research computing.
This would involve collaboration between SERC and the other Research Councils in the provision of a national resource for research computing as outlined in Option III. The Councils would work closely with the Computer Board national centres in ensuring an adequate and balanced provision of research computing, and in drawing up a properly coordinated strategic plan for the national development of research computing.
Options I and II deal with the present computing provision by SERC and address how it should be organised in the future. Options III and IV deal with broader issues of the provision of national resources for research computing and hence involve bodies other than SERC. For these reasons Options I and II are dealt with separately in sections 6.2 and 6.3 as matters that are within SERC's own remit. Options III and IV are dealt with in section 6.4 and have impact outside SERC's remit.
The current computing supported by SERC includes the items listed below and any proposals for an SERC solution must cover at least these items. In each case options on how to fund them in future are given.
The current resource used is £l.5M per year and 48MY per year, mainly at RAL. Items covered by this include: development and operation of wide area and local networking; general software support required by users from all Boards including graphics, UNIX operating system, etc; and development covering both hardware and software to enable future requirements to be anticipated. This work provides general computing support for all Boards and it is extremely difficult, if not impossible, to determine how to attribute it in a quantitative and fair way to Boards or projects. The Working Party can see no alternative to the provision of central funding if this work is to continue.
At the moment both RAL and DL provide mainframe services. The resources used for this activity are £2.1M and 60MY per year for RAL, and £l.4M and 37MY per year for DL. The RAL service is for all four Boards and for administrative computing while the DL service is mainly for work associated with scientific facilities operated by Daresbury.
The key factor in running the central mainframe service is the adoption of an effective and equitable charging scheme. An obvious option is to hand over the DL AS 7000 to the Science Board for its facility computation. It is less obvious what to do with the RAL mainframes which, if maintained to provide a service for all Boards, must be associated with a charging mechanism which generates sufficient funds to cover their replacement and up-grading, and which is self-regulating in balancing supply against demand. The Working Party believes that an appropriate charging mechanism can be evolved, and a proposal is put forward in its recommendations.
Distributed computing provided through the CCC currently includes the ICF and single user systems. The resources used for the ICF are £2.1M and 28MY per year, and for the single user systems are £O.6M and 14MY per year. These are provided through the CCC and direct payment of about £O.3M per year is made by the Boards for maintenance of the machines.
The Engineering Board uses more than 85% of the ICF and the working Party considers that the obvious and only sensible option for its future support is to return responsibility to the Engineering Board.
There are three possibilities for the support of single-user systems:
A key issue here is whether the use of single-user systems can be expected to grow substantially. It is the Working Party's view that there will be a significant growth in the use of single-user systems in all Board areas and, that while in the short term responsibility might best reside with the Engineering Board, the growth in the use of single-user systems should be monitored and the possibility of a return to central control kept under review.
In the past, administrative computing has normally been undertaken on the scientific machines as a parasitic activity. Currently the IBM 3032 has been discarded as a machine from scientific work and is being used for the development of software for administration.
It is clear that the administrative computing requirements within SERC will increase considerably over the next few years; these requirements are discussed in the administration submission in Appendix B. The four possible options for providing for administrative computing are:
In considering this matter the Working Party felt that:
We consider that computing provision, ideally, should be project-driven. That is to say Boards, having decided on their programmes of work, and having made appropriate estimates of the computing needed to carry it out, would then make the necessary provision for the computing resources which are required in order to support it. In the case of developing a large experimental facility for example, the necessary on-line computing equipment should clearly be regarded as part of the facility, as also should the computing equipment which is needed to carry out certain kinds of data reduction and the analysis of the experimental results. However, for any major programme of work, some degree of central provision of computing skills and resources will also be necessary because:
The Working Party considers that Option I should be rejected as with this option there would not be any central provision of computing skills and resources and there would be an increasing lack of co-ordination in hardware and software resources provided. This option would, in addition remove any element of strategic planning on a Council-wide basis.
We consider that Boards must realise that if central computing cannot be made cost effective, then central computing will disappear. The great advantage of an appropriately worked out version of Option II is that it would ensure that computing provision was driven by, and was responsible to, Boards' needs. An obvious long-term weakness of any straightforward implementation of this option however is that it is difficult to make within it any adequate strategic provision for the acquisition and support of very expensive supercomputers. We therefore believe that while in the short term the best way forward is a whole-hearted adoption of this option, some further form of provision must be explored to guarantee the future availability of the most advanced computing facilities to the Council's research workers.
We strongly believe that some of the research workers supported by Council will continue to need access to the very latest and most powerful computing engines in order to work effectively at the frontiers of their research fields; Appendix C gives further details. However, as discussed in section 4.2, the costs of such machines, and their staffing and maintenance, would place an unsupportable burden on an individual Board and a heavy burden on SERC. Furthermore there is an obvious need to coordinate such a provision across the activities of all the Research Councils, and hence it will be more useful to pursue Option IV than Option III. Any solution of this form must be made compatible with solutions from sections 6.2 and 6.3 above involving the other components of SERC computing.
There is a need to provide and coordinate adequate national resources for research computing including:-
Advantages would accrue from an efficient use of scarce resources and skills, and from the use of the Research Councils' well-established peer review procedures. Its disadvantages would stem from a loss of autonomy in Research Council provision. Such a loss however must be set against the great cost of procuring a supercomputer - in the region of say £10M - and of supporting its users adequately, which would involve anything from between twenty to fifty skilled staff.
There seem to be two possible options for funding such a supercomputer:
Although the Computer Board might be considered an obvious source of such provision it would have to ask a university or a Research Council to run the facility since it cannot employ staff and does not have a suitable building. If access is to be by peer review, existing structures should be used rather than the creation of a new procedure. The Computer Board is already facing a steadily increasing pressure on its financial and technical resources. The complex, distributed and highly interactive systems which universities will in future be requesting from Computer Board funds will be much more difficult to assess than the present large mainframe machines. It is worth noting that such forms of assessment and advice to universities would greatly benefit from a use of the type of skills which already exist within SERC.
We believe that serious consideration should be given to the role SERC, in collaboration with the other Research Councils, could play in the provision of advanced research computing over and above the service provided by two national centres funded by the Computer Board and the funding and organisational arrangements needed to meet this provision.
One possibility might be the creation of a third National Centre specifically for the provision of advanced research computing and subject to a peer-review process of selection. Such a centre might, for example, be run by SERC to provide advanced computing facilities for all the Research Councils. In this context, SERC appears to be the only body which currently has the expertise required to run such a machine and develop the necessary software for it.
During discussions on the future provision of supercomputers we considered whether the SERC's needs would be satisfied by access to a bureau service or by accessing large machines, for example that of the Meteorological Office, on a shared basis. Whilst we considered that the use of such a bureau service would have advantages in that funds could be switched on and off at short notice and that SERC would not have the responsibility for staffing of the supercomputer, we felt that there were considerable disadvantages which would make this option unsuitable. Such a bureau service would be relatively expensive and if SERC were the major consumer of such a resource this would probably prove more expensive than if SERC provided its own computer. Indeed if SERC were able to act as a bureau itself, the costs of computing would be substantially less. Another major disadvantage with use of such a bureau service would be that SERC would no longer have access to state-of-the-art technology and would only be able to use tried and tested machines. The prime requirement for access to computing at the forefront of technology would not be met. The use of a machine in conjunction with a body such as the Meteorological Office might be possible, but it would lead to problems over priority of access and would again require the use of proven rather than state-of-the-art computing. In either option SERC would still be required to provide a software support group.
We concluded that the use of a bureau service would not satisfy the identified needs of SERC and have not pursued this option further. However, the possibility of access to other government computing centres should not be ruled out when the requirement was for marginal use.
When the CRAY was transferred from DL to ULCC, Council agreed that, for an interim period until this Working Party had time to report, vector processing costs incurred in using the CRAY would be an allowable charge in grant applications.
As the central computing facilities at the national centres (ULCC and UMRCC) were provided essentially as an extension of local computing facilities, the charging scheme adopted by the Computer Board sets out to recover the marginal cost, i.e. the cost beyond what would otherwise have been incurred for extra local provision. The charges recovered have been determined broadly on the basis that costly facilities are adequately used while at the same time an adequate measure of control is exerted. This has been termed a market-clearing philosophy.
For the academic year 1983/84, an allocation of 20% of ULCC resources was set aside for free use by SERC grantholders. For 1984/85, the latter will pay the same rate as all other university users and universities will be expected to include the requirements of SERC grantholders in the overall bids that they make to the national centres. The Working Party considers that there is a real possibility that, if SERC were to continue to fund time on national centre machines as an alternative to making its own provision, eventually SERC grantholders could be consuming up to 50% of the available resource, thus affecting the provision for other university users and distorting the assessment of priorities. Rather than increasing the rate for all university users under its market-clearing strategy, the Computer Board might have to consider a differential rate - perhaps even full cost recovery - for SERC grantholders to reduce their overall demand or to pay for additional facilities if the demand is sustained.
If full charging were to be imposed the large amounts of money which would have to be paid would undercut the RAL mainframe service and would limit the funds available for replacement and procurement of SERC machines. There is no evidence that the requirements for mainframe computing will diminish, so such a situation would be very unsatisfactory.
We also considered that as vector processors were now provided by the Computer Board as part of the general university provision for research, to continue to pay for such charges could be seen as breaching the dual support system.
