To appear in IEEE PROCEEDINGS on Distributed Shared-Memory Multiprocessors Mar. 1999 Parallel Computing in the Commercial Marketplace. Rev 4 Nov 20 1998 1 PARALLEL COMPUTING IN THE COMMERCIAL MARKETPLACE Research and innovation at work Erik Hagersten and Greg Papadopoulos Sun Microsystems, Inc. Abstract This is a good time for parallel-computer development and research in both academia and industry. The performance improvements predicted by Moore’s Law has proven it- self quite accurate over many years. However, the doubling of processor performance every 18 months cannot keep up with the growing demand of many applications. The performance of database applications has been doubling every 9-10 months. At last, parallel-computer technology has come to play an important role in the commercial mar- ketplace. Multiprocessing has been an active research area for almost 40 years and com- mercial parallel computers have been available for more than 35 years. After getting off to a slow start, this area has now taken off . Shared-memory multiprocessors have dominated this development. This is an area that has sprung out of tireless research and numerous published breakthrough results. This article analyzes some of the reasons for the sudden acceptance of the relatively old parallel computing field, outlines the key properties of a successful parallel com- puter of the ’90s and identifies some important research areas and key technologies for the future. 1. Introduction Parallel-computer architecture is a classic research topic and has attracted good re- searchers and produced ground-breaking results the last 40 years. This technology has been applied to the commercial world over time with mixed results. Only in the last five years has this turned from being "interesting and promising" to being one of the key technologies of the world’s largest computer companies. Commercial parallel computers have evolved and matured in different distinctive waves: 1. The pioneer parallel computers (1960-2000): The first parallel computers were built from a handful of mainframe-style CPUs, connected through a switch to one common shared memory. This fairly expensive and non-scalable style of building top-of-the-line computers culminated with the CRAY-XMP architecture. This is still a viable technology provided by some mainframe vendors, but, its importance is diminishing. 2. The massively parallel processors (MPPs) (1980-1995): The cold war and its quest for computepower fueled a focused effort in academia and primarily start-up companies to build the ultimate scalable machine, dominated by massively paral- lel processors, MPPs. Many different experimental approaches were tested during this time, yet, the dominating architecture was that of the message-passing multi- computers (MMCs). The MPPs were much harder to use than anticipated. You had to really want to solve a problem with them! A lack of market volume and the end of the cold war terminated this wave abruptly. During the same time period, the