LONG-TERM GOAL

Our long-term goal is to investigate the potential of multicomputers constructed of commercial, off-the-shelf components as a replacement for conventional parallel processors in high-performance computing. By commercial, off-the-shelf components, we mean commodity workstations and personal computers connected by readily available networking fabric, e.g. DEC Alpha and Intel PC boxes connected by gigabit ethernet. To understand the impact of clusters in this area, we are developing several layers of software to support traditional supercomputing applications such as climate and ocean modeling. OBJECTIVES We hope to demonstrate the efficacy of constructing medium-scale, clustered multicomputers for a fraction of the cost of traditional supercomputers. We are building a clustered system called Centurion. Using Centurion, we will demonstrate that we can outperform a cluster of C90 machines, and at a fraction of the cost. This research can point the way to developing more effective, lower-cost computational engines for the Department of Defense. APPROACH While the original proposal was to build a 50-Gigaflop machine, the resultant machine will have more than a 200 Gigaflop peak rate. We are building this multicomputer using off-the-shelf commodit