Science: Brains that target leaky nappies with military precision: Ann Barrett looks at a campaign to encourage industry to make better use of supercomputers

Click to follow
The Independent Online
RESEARCHERS at Warwick University hope to better the biblical injunction to beat swords into plowshares by adapting a military system for identifying enemy tanks on the battlefield and turning it into a way of stopping babies' nappies leaking. According to Professor Graham Nudd, of the computer science department, 'a system that identified tanks in the grass may be able to detect the tears in babies' nappies.'

While Professor Nudd was working with the US Navy on a computer system for targeting tanks with airborne missiles, Proctor & Gamble asked him to establish a way of checking Pampers nappies for faults, using the same techniques of signal processing and analysis.

Professor Nudd is worried that British industry and commerce is failing to keep pace with electronic brainpower. In particular, he is concerned that Britain has fallen behind the United States in understanding the new generation of parallel supercomputers.

Traditional computers can hold only one 'idea' at a time - they work very fast but simply add or subtract one number after another. Parallel machines have several processors working at the same time so they can execute a number of instructions at once.

Professor Nudd is involved in a European project whose results could counter consumer resistance. Performance Evaluation of Parallel Systems (Peps) involves five groups of researchers in France, Italy and the UK, working with international standards authorities to try to introduce standards, techniques and tools into the performance assessment of parallel computers.

These supercomputers are particularly suited to tasks such as image recognition - whether it be tanks or holes in nappies - and, for more conventional business applications, to formulating and presenting information in a format that managers and executives can understand better than the dry tabulation of data from conventional computers.

British organisations appear unable to cope even with traditional machines, as demonstrated by a highly publicised series of disasters, including the collapse of the Taurus system on the Stock Exchange, the London Ambulance Service's system failure, and the waste of more than pounds 60m of taxpayers' money at Wessex Health Authority.

The use of parallel computers is further hindered by a surprisingly pervasive view in Britain that they are abstruse machines required only in scientific research. In fact, parallel machines are finding applications in industry: in designing motor cars, for example; in aircraft design and testing; in the stock market, for simulating the effects of changes in futures and commodities; and in banks, for predicting creditworthiness.

Professor Nudd's concerns are shared at official level. In 1991, a report from the Parliamentary Office of Technology and Science drew attention to the problems that arise from the lack of provision in the UK for supercomputer training and education.

The report, High Performance Computing, warned: 'Industry cannot rely on recruiting graduates who can apply a knowledge of supercomputers to solving a company's engineering and other problems; rather the company has to train its own specialists. This contributes to the low awareness, and use, of supercomputing, since the knowledge and confidence on how to take full advantage of the huge increases in computer power is lacking.'

Professor Nudd believes that Peps is vital to defining world standards in computers' performance. His survey of nearly 1,000 companies in Britain found that many would be much more confident in using a parallel computer if there were clearer criteria for assessing how any system would meet the company's needs.

Peps will look at four areas where potential users of parallel computers need to define what they want from the machines. One is simply how data is fed into the machine and how it displays the results of its calculations. Another is to develop 'benchmarks' for performance of parallel computers in critical operations. Many calculations relevant to industry - in modelling the flow of air over a new design of motor car, for example - contain operations that are repeated over and over again. If these key operations can be identified and analysed, then potential customers can ask how well a given machine performs against these basic 'benchmarks'.

Although parallel computers can carry out several computations at once, they do sometimes have a problem collating their results. As the parallel processors do their work, they need to communicate with each other, but the pathway may be blocked if one processor takes too long to perform its task. Peps will hunt out these electronic 'hot spots' and will also look at simulation and modelling of situations that are too complicated to test in simple ways.

In the United States, consultation between the computer manufacturers and potential customers is the rule before any system is selected. In Britain, a company is more likely to be pressurised to accept a system that approximates its needs and encouraged to make any necessary alterations later, when it has been tried out.

Peps will set out the basic standards that will allow companies to become more informed customers. According to Professor Nudd, the project will have succeeded 'if we do no more than tell them what questions they should ask'.