A New Programming Paradigm for Achieving High Performance Computation in Network Computing Platform

Abstract

Current parallel or distributed systems employs a large volumn of complete nodes to meet the timing requirement in solving applications of large scale. In general, the nodes are interconnected by a network of various topologies. Examples are clusters of workstations/PC 's and the computional grids. In such a platform, an appropriate strategy is needed to squeeze the computing power from the systems. In general, high performance computing can be achieved by eective resource utiliza- tion. In this paper, a novel programming paradigm is proposed for achieving eective resource utilization in network computing platform. We use two parameters to model the platform. Based on the model, a function is proposed to capture the programming feature on the platform. The function is used to group data items into messages and to schedule the messages to their destination site. By choosing a grouping function appropriately, we can maximize the utilization of the computational resources in the system. Finally, matrix multiplication and LU decomposition are used as examples to illustrate the usefulness of our programming paradigm. In this paper, we has shown that computational resources can be eectively utilized by employing our programming paradigm.