Å®ÉúСÊÓƵ

Experts in parallel computing, in which anything from a handful to a
thousand or more chips work collectively, met in Germany this week and in
the US last week in separate bids to sort out what is probably the biggest
problem facing them. Programs written for one parallel computer will not
work on any other type.

Similar problems have always afflicted new types of computer. Personal
computers, for example, only became really successful when IBM’s PC, with
Microsoft’s DOS software for handling the basic instructions, came along.
It became the standard, and many companies produced hardware and software
which was compatible with it. This has not happened yet in the parallel
computer industry.

Tony Hey of the Parallel Applications Centre at the University of Southampton,
says there are two important areas of dissent. One is how the computers
pass messages – data or instructions – between the processors. Genesis,
a project funded under the European Community’s Esprit programme, has already
produced software which can run on the different computers made by Intel
and NCube of the US, Meiko in Britain and Parsytec of Germany.

The other problem area is memory. ‘We could have lots of processors
wanting to read or write to the same part of memory at once,’ says Hey.
If this is not catered for there can be anarchy, and the program may make
mistakes. But exactly how the memory is controlled varies between computers.

Hey is optimistic that the meetings will have got some of the key companies
talking, including Fujitsu, Cray and IBM which all sent representatives.
‘I’d like to think something will be decided within a year,’ he says. He
believes that Europe, which has a large number of small parallel computing
companies, is in a position to influence the standards, though he says ‘I
suspect most of the activity will be driven by the US.’

Greg Wilson of the University of Edinburgh’s Parallel Computing Centre
is less hopeful. ‘I don’t see the willingness to compromise that I’d like
to. Until IBM or Fujitsu says, ‘here’s the standard’, I don’t think we’ll
see much.’

But there are signs of the parallel processing industry settling down.
‘In the last two to three years we’ve seen people using standard processing
chips, being connected in regular topologies,’ says Wilson. And independent
software companies are starting to write parallel programs – quite a risk
unless the hardware is beginning to standardise.