Tempting "big-science" prizes such as the Square Kilometre Array (SKA) radio telescope, for which Australia and New Zealand are bidding against other nations, will demand expertise in parallel processing, to deal with the massive flow of data to come from its observations. Other data-intensive disciplines at the forefront of science, such as genomics and climate modelling also demand parallel processing techniques in software.

The work of Erdödy and his colleagues at Oamaru-based Open Parallel has attracted the attention of Intel. James Reinders, marketing and business director for Intel's software developer products division, attended a Wellington meeting last week at which Open Parallel discussed the need for expertise in parallel processing.

The future of parallelism is the future of computing, Reinders says; around 2005, Intel stopped increasing clock speeds on its simple processors, facing problems of increasing power demand and heat dissipation. Instead, it used increased transistor density to put multiple cores on one chip, genuinely enabling the processor to do more than one thing at once, in contrast to the illusory concurrent activity of multitasking in the operating system.

Reinders is impressed with the work Open Parallel researchers and others have been doing to introduce parallel-processing commands to the PHP and Perl languages and thereby the WordPress blogging environment and other open-source products, using Intel's Threading Building Blocks (TBB) library. The projects implemented TBB commands into HipHop, a source-code transformer developed at Facebook. HipHop translates PHP into optimised C++, which it then executes in a particularly efficient way. Facebook open-sourced the HipHop code last year.

The Open Parallel team "made slight modifications to WordPress, to use a TBB-enabled HipHop they created -- and they had very impressive results," Reinders reports. "The bigger surprise was the huge drop in memory footprint." A task divided into several simultaneous sub-tasks can take up less memory than the same task as a single entity, he says.