~9I~o------SPRING80QKS------~N~A~T~U~R~E_v~o~L~.3~26~3~0_A_PR~I~L_l9~87 machines commercially available are the Engines of change Cray and Control Data Corporation John Forty ranges, which exploit the concepts of multiple central processing and architec­ tures which permit pipelining of a series of Supercomputers and Their Use. By similar processing steps which can thereby Christopher Lazou. Clarendon: 1986. be processed simultaneously, and conse­ Pp.227. £22.50, $45. quently increase the speed of calculation. The latest Crays incorporate up to four THE generally accepted definition of a powerful central processors so that this supercomputer is a general purpose com­ vector processing of data can be per­ puter which is the most powerful comput­ formed in parallel. Machines such as the ing engine available at any given time. In ICL DAP (distributed array processor), principle, therefore, there can only be one consisting of a very large number of simple such machine but in practice a range of processing units working in parallel, are supercomputers is available and the extremely effective in carrying out large choice depends on performance for the scale computations where there is a large particular application. Inevitably the pace number of similar tasks to be performed of advance in computer technology is such simultaneously. Computing engines of that no machine can hold the title for more this kind are orders of magnitude more than a few years; thus, in its day, the powerful than the conventional main­ Ferranti Atlas was a supercomputer but frame. was soon overtaken by the very powerful A useful measure of progress over the mainframes of the 1960s such as the IBM past 20 years can be obtained by compar­ 360/195 and the CDC 7600. ing peak performances, today's multi­ In the past decade there have been processor machines, such as the 4- Cray and his creation - Seymour Cray with enormous advances in computing power, processor Cray X-MP48, being several the Cray-2 machine, announced in June 1985. defined loosely in terms of the ability to thousand times faster than the Atlas. The basic concepts of supercomputers, deal with the very large scale computation Advances in hardware technology have the principles underlying their architec­ needed to solve complex problems, also resulted in a reduction of the cost per ture and the developments in semicon­ usually of a scientific nature. These have computational step by a similar factor. ductor technology which, of course, are come about largely as a result of the The result is an improvement in per­ fundamental to their performance, are set development of new forms of computer formance which is the equivalent of out in the first four chapters. The treat­ architecture. Currently the most powerful travelling at ten times the speed of Con­ ment is inevitably superficial, but the text corde compared with walking pace, for and well-chosen illustrations make it easy the same cost! to grasp the essential features. There are Computer design and engineering is also some useful chapters on scientific continually evolving, and it might be computational methods, including scalar thought that any account of super­ and vector techniques, and optimal computers as we know them today could vectorization of programs. The reader is be of little permanent value. That may be introduced to the concept of performance so, but Christopher Lazou's little book in terms of hardware and software para­ does serve a useful purpose. It draws meters such as the number of parallel attention to the emerging scientific processors, degree of vectorization and method of computation which is now seen the fraction of work that can be processed by scientists to be a valid and technically in parallel. As its title declares, the book feasible approach to solving problems also covers the uses of supercomputers (usually but not always of a scientific and a number of examples are described, nature) which are so complex that the drawn largely from the author's experi­ traditional approaches of experiment or ence with the Cray 1 (the supercomputer analytical investigation are impractical. of the late 1970s) at the University of Supercomputers have made it possible to London Computer Centre. Looking for­ study the behaviour of the Earth's ward, there is a useful survey at the end of atmosphere and oceans in great detail, the book of new supercomputers known which will lead to a better understanding to be in the pipeline and of future develop­ of climate, weather and so on. We can ments that should be made possible by produce computer models of whole oil advances in technology. reservoirs to optimize output. And air­ Lazou writes in a fluent, informative craft designers are able to design parts of, manner. Although the subject is not or even whole aeroplanes, and simulate explored in depth, the book provides a the aerodynamic performance over a wide good introduction for those who are range of flight conditions with such real­ already computer literate and wish to ism that the time-consuming and expen­ know more about these developments at sive process of wind-tunnel modelling is the forefront of computer technology. 0 no longer necessary. An important aim of John Forty is Vice-Chancellor and Principal of the author has been to draw attention to the University of Stirling, Stirling FK9 4LA, the strategic importance of the super­ UK. He is the author of Future Facilities for Advanced Research Computing, the report of a computer to the many areas of scientific Joint Working Party of the Advisory Board for research and engineering design in which the Research Councils, the Computer Board Reader Service No.7 it is now being applied. and the University Grants Committee.