Network-based scientific computing

Abstract

Rapid advances in modern networking technologies and commodity high performance computing systems are leading the field of computing in a new paradigm referred to as network-based computing (NC). This paradigm views a large number of geographically distributed computer resources such as PCs, workstations, Symmetric Multi-processors (SMP) and Massively Parallel Processing (MPP) systems connected through a high speed network as a single meta-computer or computational grid [2]. In this paper, we focus on the Internet (WAN) and Intranet (LAN) based computational grids and their ability to support scalable and robust "deep" computing. We present various implementations of the NC paradigm using commodity and customized software in the context of the existing PELLPACK problem solving environment [8], the ITPACK library as it has been implemented in the PELLPACK system, and a muti-physics application for the design of gas turbine engines [11]. Through this study we attempt to assess the feasibility and efficiency of several NC paradigms for scientific applications utilizing existing middleware. © 2001 by Springer Science+Business Media New York.