Portable and Scalable Parallel Applications with VCluster

Message passing based parallel programming paradigm and associated libraries such as MPI and PVM have proven its novelty and efficiency by successful applications in many diverse areas ranging from scientific computation, simulation, graphics, machine learning, to data mining. Those tools have been de facto parallel programming libraries for cluster computing where a cluster consists of usually homogeneous uni-processor machines. Recent advances in processor technologies made it possible to build a cluster of multi-processor machines more economically. In addition, a new programming language, Java, and its associated technologies opened a door to flourish of new and more efficient development of distributed computing software. Virtual Cluster Computing (VCluster) library is a portable parallel runtime system we have developed to support parallel programs that run on the cluster of heterogeneous uni- and multi-processors. In this paper, we briefly introduce the features of VCluster system and describe how parallel programs can be developed efficiently with VCluster. We also present the experimental results showing the performance of the developed system on two exemplary applications, heat diffusion and back propagation neural network, and compare them with other systems