n-Cube model for cluster computing and its evaluation

Abstract

Cluster systems are widely used in modern high performance computing. With the rapidly increasing of parallel algorithms, it is an open problem to analyze and evaluate whether they take good advantage of the computing and network resources of clusters.[1-3] We present a novel mathematic model(n-Cube Model for Cluster Computing) that epitomizes the algorithms commonly used on clusters and evaluate this model using Stochastic Petri Nets (SPN). The state space of our model's SPN is also discussed formally. Finally, we take MM5(the FifthGeneration Model) as a case and the comparative performance analysis shows the immense vitality of the model. © Springer-Verlag Berlin Heidelberg 2007.