High-performance, low-cost grid computing through community cooperation

Computationally complex grid-based applications can require expensive high-end computer equipment, forming an effective barrier to many small companies. However, if a grid-based problem can be decomposed into many smaller problems, each solvable on a moderately powerful consumer PC, such PCs could in aggregate provide a solution to the performance barrier. Thus, the problem becomes one of procurement; where can so many PCs be obtained? In this paper, we describe a low-cost, highperformance grid developed by Noregon Systems using excess computational power from the computers of the local school system. We present a case study of how Noregon Systems and Targacept, Inc. used this grid for state-of-the-art molecular simulation. We demonstrate the feasibility and scalability of such a grid for other high-granularity calculations, and present software architecture capable of supporting multiple grid middleware tools and multiple applications.