New development hub for parallel programming

Commercially they're competitors, but on campus it's all hearts and flowers between chip manufacturers AMD, Intel and Nvidia and systems houses Sun Microsystems, HP and IBM. This Friday, in conjunction with Stanford University, they plan to launch the Pervasive Parallelism Lab. Under the leadership of Stanford professor Kunle Olukotun, the lab, which has a budget of six million dollars for an initial three years, will look for novel ways of simplifying the complexities posed by parallel programming – new operating systems, compliers and processor architectures. Olukotun already has plenty of experience in this field, having previously headed the Hydra project which developed single-chip multi-processor technology, the results of which were picked up by Sun for its Niagara processor.

The researchers are in competition with their colleagues at the University of California at Berkeley, who announced a similar alliance with Intel and Microsoft, the Universal Parallel Computing Research Center, in collaboration with the University of Illinois just last month. The approach being taken by the Berkeley researchers headed by David Patterson is somewhat different. They have divided the problems posed by parallel programming into seven classes, each of which is to be solved after its own fashion.

Both universities have been shaped by outstanding computer scientists over the last several decades – John Hennessy, currently President of the University, at Stanford and David Patterson at Berkeley. Between them the two have written two of the most important books on computer architecture ever published – "Computer Architecture: A Quantitative Approach" and "Computer, Organization & Design". Hennessy was a co-founder of MIPS, whilst Patterson laid the foundations of the SPARC architecture, which was later taken up by Sun – itself a Stanford University spin off.

Both are now working intensively on parallel structures. An interview with Hennessy and Patterson chaired by Kunle Olukotun, the new head of the Pervasive Parallelism Lab, from a little over a year ago – in part as an mp3 – offers a nice overview of their areas of overlap and divergence.

(trk)