A One Language Fits All Approach To Parallel Processing

Computers stuffed with dozens of microprocessors can outperform the fastest "super" from Cray Research Inc. That's because all those chips simultaneously pick away at pieces of the job. But to wring these speeds from a parallel computer, programs usually need to be rewritten to meticulously control the flow of data to and from each chip. That's a formidable task, so not many rewrites have been tried.

Researchers at the European Computer-Industry Research Center (ECRC) in Munich believe they've found a way around this software bottleneck, using something borrowed and something new. First is a programming language that doesn't require instructions to be doled out in rigid sequence, as Fortran and most others do. With so-called logic programming, "you can shuffle instructions like a pack of cards," says Michael W. Freeston, head of technology transfer at ECRC.

That removes the need to tightly control the data going to the microprocessors. But there's still the problem of keeping track of the piecemeal results coming back. For that, ECRC has developed a special data base featuring "data persistence." It keeps tabs on the pieces until the final answer can be assembled. London's International Computers Ltd. is already using the technology to help manage Hong Kong's busy port.

Before it's here, it's on the Bloomberg Terminal.