At the end of 1984 the Advisory Board for Research Councils, the Computer Board and the University Grants Committee set up a joint Working Party 'to consider and report on the likely needs for advanced research computing and on the various options open to the university and research council community for acquiring, operating and providing access to the necessary services'. The Working Party was chaired by Professor A J Forty, then of Warwick University. Its report was published in August 1985 and received wide coverage in the press. Dr Brian Davies reviews the report and subsequent developments.
The Working Party found a strong case for the provision of new advanced computing facilities, and believed that there were opportunities for advancing knowledge and understanding by computational methods in almost every branch of science. In effect the greatly increased computing power offered by present day technology has opened up a new dimension to the experimental method in scientific investigation. In many areas of science, the physical laws governing the behaviour of matter and materials are well known, but the application of these laws to complex systems which are of pressing scientific interest results in equations which require enormous computing power to solve. Examples of the areas which could benefit from this kind of computing include aircraft design, the evolution of galaxies, the chemical reactions of molecules, the physics of liquids and solids, large-scale integrated circuit design, the circulation of the atmosphere and oceans, and the design and operation of new pharmaceutical products.
The Working Party recommended that ABRC, the Computer Board and UGC should secure and allocate funds for a national facility for advanced research computing. This should comprise:
Given the scale of the recommendations, there has been remarkable progress towards their implementation during the past year. In February 19S6.8the ABRC. Computer Board and UGC agreed on a funding package which would provide a Cray X-MP/48 supercomputer to be installed at the Atlas Centre at the Rutherford Appleton Laboratory, and a less powerful Cray 1S supercomputer to be installed at the University of London Computer Centre. Both these machines have now been delivered. The Cray 1S at ULCC came into service in the Autumn of 1986. The X-MP/48, delivered in December 1986, should by now be running its first user jobs.
In addition the Computer Board has set aside £5 million to be spent over three years on enhancements to networking facilities, and reviews are being made within the Computer Board and SERC of the need for and possibility of providing distributed facilities.
The Cray X-MP/48 is one of the most powerful computers it is possible to obtain. In principle it is capable of performing about a thousand million arithmetic operations per second. In practice it is unlikely to attain this extraordinary level of performance because it would mean making every individual section of the computer perform simultaneously flat out, and although programs can be written to do this they tend to have little relationship to real life applications. Nevertheless, by careful tuning and optimisation of existing scientific programs it is possible to attain a substantial fraction of this theoretical performance for real problems and this is very considerably more than can be obtained from conventional mainframe computers.
The computer achieves its high performance in three main ways. First, it is constructed from very fast components and all possible steps have been taken in its design to reduce the transit time for electrical signals to move from one point to another. The computer is therefore very compact - it occupies only six square metres of floor space. Its compactness leads to formidable cooling problems and the machine is cooled by Freon refrigerant rather than by the air or water used by more conventional computers.
Secondly, the machine contains special hardware designed to handle vectors (strings of numbers) in much the same way as a conventional computer handles individual numbers. This can provide large gains in performance when dealing with repetitive operations on arrays of numbers, as for example in doing the kind of matrix algebra which features prominently in solving sets of equations.
Thirdly, the machine contains many functional units which can operate simultaneously, and clearly, the more one can exploit this parallelism, the greater the overall performance that can he delivered.
The above features apply to any Cray computer. The X-MP/48 model contains a further element of large-scale parallelism in that it consists effectively of four complete Cray computers (processors) in one box. It is possible to run the four processors almost as four independent machines or, by adding extra instructions to users' programs, one can make all four processors cooperate on a single job.
The X-MP/48 has 8 million 64-bit words of memory and a 32 million word solid state device (SSD) to be used for data files which are required frequently by user programs. Use of the SSD can make it feasible to perform certain types of calculations which might otherwise be impractical if one had access only to traditional, and much slower, disk drives for such data storage.
The Cray also has its own conventional disk drive units and it will have access to other facilities which are now available on the IBM mainframe computing facility at RAL. This IBM system will act as a 'front end' to the Cray. It will provide the route into the Cray from JANET; it will be the machine on which much of the preparatory work is done by users before they submit their jobs to the Cray; and it will provide large-scale storage facilities for long-term data storage. In time, further front-end machines providing somewhat different facilities may be added as funds become available.
