Computational services in the UK were given a major boost in 1980 when the first Cray vector processor was installed at the SERC Daresbury Laboratory for the scientific community. The Collaborative Computational Projects (CCPs), bringing together active groups in a variety of disciplines, were quick to exploit the new facility and to give the UK an edge over workers in other countries. However, the transfer of the Cray IS from Daresbury to the ULCC in 1982 proved to be a major disruption of many of these programmes. At the same time, other countries, notably Germany, Italy, Japan and the USA had moved forward with generous provision of supercomputers for academic users. It appeared that Britain had lost the initiative.
The Joint ABRC, UGC and Computer Board Working Party was set up in 1984 to investigate the need and recommend provision for advanced research computing in the UK. The Working Party formed two sub-committees, one to examine the scientific case and the other to study the technical options that were available or expected to become available. The Working Party was asked to consider the likely cost of any new provision and how it might be met. It was also asked to give advice on how such facilities should be administered. The outcome was the Report published in July 1985.
The main recommendations of the Report were:
Most of these recommendations have now been acted upon The Computer Board has been allocated a sum of £5M from the DES for the enhancement of JANET over the next 3 years. The Computer Board has also purchased the Harwell Cray IS and installed it at ULCC. The front end computers at ULCC and UMRCC have been enhanced and it is intended to make the user interfaces at the three centres as similar as possible. A Cray X-MP/48 has been ordered, with joint funding from the ABRC, UGC and Computer Board, and delivery is expected at the end of November 1986 - almost a year earlier than was anticipated. The Computer Board and SERC have set up a collaborative group to coordinate the provision of DAPs. FPS processors and other special purpose machines over the next few years to provide the distributed facilities. Other research councils have indicated their willingness to make such provision for their communities.
The outcome of the Forty Report has therefore been a wide acceptance of our recommendations and a most encouraging rapid response by special funding of some £30M over a 4 year period. Decisions have yet to be taken about the third phase of supercomputer provision in 1990 and also the very important recommendation in the Report on the need for interdisciplinary fellowships. It now remains for the academic community to exploit the new facilities to real advantage.
The Cray X-MP/48 will have the following configuration:
The funds required to run the service and to buy the computer are based on a formal acceptance of the machine in April 1987. In practice the computer is expected to be delivered around the end of 1986 and it is hoped that the first user jobs can be run towards the end of January 1987.
The funding arrangements assume that the Cray service is being added to an existing substantial mainframe computing service organisation. The additional manpower provided to run the Cray service is therefore small - 4 staff, and the additional finance covers only the Cray maintenance bill, the machine's power consumption, its software to link to front end facilities and the costs of the 4 staff. Separate provision is being made for the specialist scientific support, the need for which the Forty report stressed so strongly. The existing IBM compatible mainframes at RAL will provide front-end facilities for the Cray as well as continuing to provide the present scientific computing services.
The two mainframes (IBM 308IK and Atlas 10) are rated at 16 and 15 mips respectively. Between them they have access to 40 Gbytes of disk storage, a mass storage device with 110 Gbytes, 16 magnetic tape drives and a variety of other peripheral devices.
The Cray will be connected to the IBM 308IK. This machine is connected to the Joint Academic Network JANET and is the computer on which interactive technical services are supported. The Forty report recommended that Cray users should use IBM's VM/CMS terminal system for front end work, and this is the system already in use on the 3081K.
The report also recommended VAX/VMS and Unix front end facilities to complement IBM based front ending. Steps are being taken to enable facilities of this type to be connected as funds become available to buy them. So far such funds have not been provided.
Present ideas on dataset management for the Cray are that scratch datasets of up to 0.5 Mblocks (32 Mwords) will be allowed initially on Cray disks. Permanent datasets of up to 100K blocks will be allowed routinely. Datasets will be backed up onto users' CMS minidisks. Any Cray datasets that have been unused for 7 days will be migrated to users' CMS minidisks. After a further period of inactivity they will be migrated to mass storage and scratched after one year of inactivity. Initially data from magnetic tape will be transferred to the Cray via CMS minidisks. Direct access to magnetic tape from the Cray will subsequently be provided, but there may be some restriction on usage.
The software libraries installed on the Cray X-MP 48 computer will reflect the requirements of users. It is proposed that the initial set should comprise IMSL. NAG including NAG graphics, GHOST. GINO and GKS. Some application packages will be mounted. such as those developed by the Collaborative Computational Projects. It is expected that requirements for others will emerge as users seek services on the computer.
The Forty report proposed that a user support group should be established at the Atlas Centre predominately by fixed term appointments involving research council, university or industrial partners. In addition it was proposed that the support provided by the SERC Collaborative Computational Projects at the Daresbury Laboratory should be coordinated with the user support provided at the Atlas Centre. Finally it was recommended that a Fellowship Scheme for Computational Science and an Overseas Exchange Fellowship scheme should be established.
Although the ABRC was not able to provide funds for the user support group or for the Fellowship Scheme, proposals for establishing a small scientific support group at the Atlas Centre have been made. In addition the Collaborative Computational Projects and international collaborations are expected to be vital in the support of the Cray X-MP. In this contribution these proposals are described.
It is proposed to set up a small scientific support group at the Atlas Centre. This will consist of the following:-
In this way it is hoped to bring together a scientific support group consisting of about 10 people who will work in the main areas of research carried out on the Cray X-MP.
The Collaborative Computational Projects (CCPs) each bring together about 10-12 university groups to develop, collect and maintain applications software and to exchange information on theory, algorithms and programs in selected areas. The projects are supported by permanent Daresbury Laboratory staff and research assistants and are allocated a travel budget for workshops and discussion meetings as well as computer time. At present these projects are coordinated by the Theory and Computational Science Division at the Daresbury Laboratory. In the future the work of the CCPs will be orientated to the use of the Cray X-MP.
The CCPs which are at present supported and the Chairmen of the working groups supervising the projects are:
Discussion meetings are now planned to consider new projects in other areas of science and engineering. Further information concerning the present and future projects can be obtained from Dr J E Inglesfield at the Daresbury Laboratory.
A number of the CCPs have overseas members on their working groups. In addition SERC supports international collaboration through the Centre European de Calcul Atomique et Moleculaire (CECAM) in Orsay France. Discussions have also been held with the proposed Centre European de Recherche et de Formation Avancee en Calcul Scientifique (CERFACS) in Toulouse France.
In the future it is hoped that links will be established with overseas supercomputer centres in the USA, West Europe and Japan.
Although the final form of the scientific support provided for the Cray X-MP has yet to be decided it is regarded as important that there should be a strong scientific group at the Atlas Centre coordinated with the Daresbury Laboratory group and with overseas groups.
Academic users currently have access to the ULCC and UMRCC machines by way of block allocations made by these centres to individual universities. Each university is responsible for making its own bid (possibly coordinated through a regional centre in the case of ULCC) and also for paying for its own allocation. Some of this access is supported by research council grants following peer review.
The new arrangements for advanced research computing introduce changes in the wider use of peer review for proposals. The advanced research computing facilities will be funded centrally and, apart from exploratory use, access will be by peer review. The peer review bodies are the various subject committees of the Research Councils and the British Academy. SERC operates three grant rounds (closing dates: 15 September, 15 December, 1 April) but other bodies meet with different frequencies. Each will generate prioritised lists of peer reviewed projects which will be fitted within their allocation if overall demand exceeds capacity. It will be necessary to define allocations of the resources for the various disciplines and research councils so as to guarantee proportional access to the users applying through the various routes. Management of the distribution of resources will be important for a fully loaded machine.
The introduction of these new arrangements is still subject to a few uncertainties. At the time of writing it is likely that ULCC will be able to accommodate new "peer reviewed users" from February 1987. It is hoped that the Atlas Centre service will also start in February 1987. Peer reviewed access to the UMRCC service will be possible from August 1987 i.e. at the start of the new allocation year.
At the moment plans are being drawn up to unify the method of application to all three centres. A National Supercomputer form (NS1) is being designed. The existing application form for SERC computing resources (well known to many Atlas Centre users as form AL54) is in temporary use and is available from Resource Management, Atlas Centre, RAL. Eventually copies of the new form will be available from your local computing centre, with whom you should always discuss your requirements in the first instance.
In his autumn statement for 1985, the Chancellor announced additional funding for the Computer Board to support advanced networking. The initiative covers enhancement of both the JANET wide area network and the local facilities on each University Campus. The total budget available is five million pounds over three years, of which about thirty percent will be spent on the wide area.
The aim of the upgrade programme is to increase the performance and eligibility of communication across JANET. This is being achieved by installation of higher speed switches and faster trunk links between the major switches, and by upgrades to site access links. In the first phase, a box of 2 megabit circuits will be set up between London, Manchester, Dares-bury and the Atlas Centre at Chilton. Multiplexing equipment will be used to provide primary and reserve quarter megabyte channels from each of these switches to the other three. To exploit this fast central core, the access links from individual sites will be upgraded when traffic warrants it; the target is to provide at least 48 kilobits per second to all sites by the end of the three year programme.
Access to major facilities such as the Cray plays a large part in the planning of the upgrades. It is expected that direct access to these centres will be provided at speeds higher than for other sites; where necessary, connection will be made to more than one JANET switch to spread the load and improve response.
When the upgrade programme is complete, users should see consistently low network transit times, giving predictable response, and improved throughput for bulk data transfer. Rates of 48 kbps into or out of a typical site should be achieved, and will generally be available to individual users.
The switch upgrades envisaged have now been performed, and the high speed link and multiplexing equipment have been ordered for delivery early in 1987. The Network Executive is planning a phased schedule of testing and gradual incorporation of links into the production network aimed at the introduction of all the new equipment into full service by the end of next summer.
"What kind of activities is it sensible to perform across JANET?" This question may be posed by potential users of the Cray X-MP. An answer will interest them. As a reasonable guideline, projects will not be greatly affected by remote operation if the time needed to transfer data across the network is less than 25% of the time needed to generate the data. If a calculation takes several hours to perform, then a transmission delay of tens of minutes is no great burden. If large amounts of data are produced in seconds, transmission times in minutes make remote access impractical. It should be noted here, however, that the production time of concern is the full service time of the remote resource, and that it is the queueing time of large shared systems which often determines the kind of application interactions which can be supported.
However, performance does not become an issue until one has the necessary access facilities. Considerable effort is being made by all three national supercomputer centres to ensure that the full set of access protocols used in the academic community are available. This basic set supports job transfer and manipulation, interactive access in both line and screen oriented styles, file transfer and electronic mail. It is hoped that the result of these activities will be straightforward access to the Cray X-MP and to the other national supercomputing facilities.
