A Distributed Object Based Framework for Parallel ComputationsÃ

Abstract: The computational and compositional features are very important while constructing parallel software for the workstation clusters. However, lack of suitable supporting environment for parallel software development makes most existing distributed parallel software systems very weak in these two aspects, especially in the compositional feature. In this paper, a distributed object based framework for parallel computation is proposed. The goal of the framework is to achieve high efficiency for parallel computing, to construct a mechanism to encapsulate and reuse parallel programs, and to guarantee load balancing and fault tolerance. The framework is a four-layer model that includes an object-group layer and a mobile object layer. The experimental results verify the efficiency of the scheme. Key words: distributed object; mobile object; parallel computing; workstation clusters; framework Parallel computing has become the natural medium for solving a wide variety of computationally intensive problems in a very fast and efficient manner. Currently, more and more distributed software systems base their platforms on heterogeneous connected workstation clusters. However, the inherent complexity of many parallel computing applications makes development using existing parallel programming paradigms both time-consuming and error-prone. Recently, many parallel computing environments such as MPI[1] and PVM[2] have been proposed, but thes