The scientific workstation of the future may be a pile of PCs

A rapid convergence of computer technologies from two historically disparate markets may dramatically alter the way scientific single-user systems are implemented in the near future. Once relegated to the menial tasks of games, word processing, and spreadsheets, PCs and their underlying component technologies are rapidly overtaking what had been an exclusive workstation market. The healthy lead in performance and functionality held by workstations for over a decade has since dissipated to the point that the distinction between the two classes of machines may soon be extinct. What has not changed is the cost benefits of mass-market commodity PC subsystems with a scale of manufacturing in the millions of units per year rather than tens of thousands typical for workstation vendors. A second emerging factor in the evolution of desktop systems is incorporation of parallel processing—the integration of multiple microprocessors within a single ensemble, first introduced at a modest scale by the workstation vendors a few years ago. PC-based systems with two processors have been available for the last two years and up to four processors in a single system may be available as early as 1997. But a much higher degree of parallelism is feasible for PCs within the cost constraints of typical high-end scientific work- stations. Significantly, exploitation of this new potential is possible now and need not wait for some future product release. Methods for applying a pile of PCs (PopC, pronounced “pop-see”) to a working environment—once restricted to the finest workstations—has been developed at the NASA Goddard Space Flight Center. The Beowulf project has shown that in the world of cheaper-better-faster, PC-derived technology combined in parallel structures may yield superior operational characteristics at competitive cost through the PopC approach