High performance computing: Clusters, constellations, MPPs, and future directions

Last year's paper by Bell and Gray [1] examined past trends in high performance computing and asserted likely future directions based on market forces. While many of the insights drawn from this perspective have merit and suggest elements governing likely future directions for HPC, there are a number of points put forth that we feel require further discussion and, in certain cases, suggest alternative, more likely views. One area of concern relates to the nature and use of key terms to describe and distinguish among classes of high end computing systems, in particular the authors use of ''cluster'' to relate to essentially all parallel computers derived through the integration of replicated components. The taxonomy implicit in their previous paper, while arguable and supported by some elements of our community, fails to provide the essential semantic discrimination critical to the effectiveness of descriptive terms as tools in managing the conceptual space of consideration. In this paper, we present a perspective that retains the descriptive richness while providing a unifying framework. A second area of discourse that calls for additional commentary is the likely future path of system evolution that will lead to effective and affordable Petaflops-scale computing including the future role of computer centers as facilities for supporting high performance computing environments. This paper addresses the key issues of taxonomy, future directions towards Petaflops computing, and the important role of computer centers in the 21st century.