Development of Effective Tools for Cluster Computing and Applications to Problems in Materials Chemistry and Physics

Iowa State University, Ames IA

Investigators

Abstract

With support from the Major Research Instrumentation (MRI) and Chemistry Research Instrumentation and Facilities (CRIF) Programs, and the Division of Experimental and Integrative Activities (EIA), Prof. Mark S. Gordon of Iowa State University, in collaboration with IBM, Myricom and Packet Engines, will design and build a cluster computer. This computer will look and perform logically like a moderately sized MPP system, but at a fraction of the cost. Key to the development of a successful machine is optimal inter-node communications, and the PI will utilize a) the expertise of the Iowa State University Scalable Computing Lab together with the theoretical chemistry and condensed matter physics groups; b) access to AIX source code via his partnership with IBM; and c) optimization of communications technology via interactions with Myricom and Packet Engines. Gordon will focus on optimization of throughput, reduction of latency, scalability of real applications beyond 64 nodes, cluster heterogeneity, distributed file systems, development of more effective message passing tools, system management and scheduling, and applications in materials science. Computational science and engineering is positioned to play a leading role in solving grand challenge problems that we face in the next decade. These important problems include the design of new materials and catalysts with desirable properties, the elucidation of biological processes, the search for the origin of life, and the development of viable methods for environmental remediation. Solution of such compelling and complex problems requires not only state-of-the-art computational hardware, but also the development of the necessary models and algorithms to take optimal advantage of modern computers. This work will lead to the development of a cluster environment that will enable grand challenge applications that require scalable high-performance computing.