Performance analysis of grid architecture via queueing theory

With the remarkable development of grid technologies, many grid architectures have been proposed which have evolved greatly in recent years. One of the major challenges for grid technologies is to build the scientific and technological base for share, collaboration, large- scale distributed systems. Theories and models of grid architectures are crucial to this attempt as well as to underpinning the foundations for constructing grid systems working effectively. On the basis of a coarse-grain classification of grid applications, we present a novel taxonomy of grid architectures: interaction-intensive architecture and computation- intensive architecture. In this paper, two new grid performance metrics are proposed to measure performance of grid systems. Then, we model grid architectures mathematically via queueing system M/M/1 and M/M/1/K; mean waiting time and total spending time are given. Furthermore, we investigate the performance of two grid architectures and compare them with each other; meanwhile a series of simulations are carried out to demonstrate the analysis results. Finally, we obtain some scientific principles guiding the grid architecture design and optimization.