This year marks the tenth anniversary of the first Beowulf PC cluster and the end of an extraordinary decade of expansion in the capabilities of commodity clusters in general. During this historic period, clusters have advanced from modest experimental systems delivering a few hundred Megaflops to the dominant class of high-end computing systems today delivering on the order of ten Teraflops or more. The majority of systems on the Top 500 list measured by the Linpack Benchmark are commodity clusters (including Constellations). From a few academic laboratories in the early 1990s commodity clusters are now nearly everywhere including National labs and centers, university machine rooms, industrial facilities, and general commercial services providers. But even now, emerging trends in component technology, processor architecture, interconnect networks, and programming and resource management software promises dramatic improvements in performance, performance to cost, power, and footprint to deliver unprecedented capability of multiple Petaflops a decade hence. This keynote address will explore those trends that will drive commodity clusters towards a realm of performance that will exceed the total aggregate throughput of all the machines on the Top 500 list today. In this context, key research issues will be discussed that are likely to provide the critical enablers to power this exciting next decade of advance in commodity cluster computing.
Thomas Sterling is a Principal Scientist at the NASA Jet Propulsion Laboratory Information Technology and Software Sciences Division and a Faculty Associate at the California Institute of Technology Center for Advanced Computing Research. He received his Ph.D as a Hertz Fellow from MIT in 1984. In 1994, he initiated the Beowulf Project at the NASA Goddard Space Flight Center which led to the wide spread use of low cost Linux-based PC clusters for technical computing. He is the co-author of two books on Beowulf clusters published by MIT Press: How to Build a Beowulf and Beowulf Cluster Computing with Linux. He and his colleagues won the Gordon Bell Prize for this work in 1997. Thomas Sterling was a leader of the National Petaflops Initiative, an open community exploration of the complex issues related to Petaflops scale computing. He co-authored the MIT Press book: Enabling Technologies for Petaflops Computing and was Principal Investigator of the HTMT Project, a detailed point design study of a revolutionary computer concept incorporating innovations in technology, architecture, and programming, and involving over 70 researchers in a dozen institutions. Thomas Sterling is a researcher in advanced processor in memory (PIM) architecture and is the PI of the NASA Gilgamesh Project to develop the MIND PIM chip for scalable fault-tolerant real-time space-borne applications. His current active research endeavors include: 1) the Cray Cascade Project to develop a Petaflops computer by the end of this decade under sponsorship of the DARPA High Productivity Computing Systems Program, 2) the Argonne-led Advanced Programming Models Project under DOE sponsorship, and 3) the Sandia-led Fast-OS Project under DOE sponsorship. Dr. Sterling holds six patents.