Analysis of prioritized cluster-based crossbar systems

This paper analyses the performance of cluster-based crossbar systems under rite assumption that processors are organised in a priority hierarchy. The effective memory bandwidth is chosen as a main measure. Two situations are considered: (1) a processor-has a few memory modules; (2) the requests from processors are uniformly distributed among all memory modules. Analytical results are validated by comparison with simulation data. We also discuss the effect of the probability, which is that a processor generates a request in a cycle, on the performance of this system. In addition, we compare the bandwidth of prioritised cluster-based crossbar systems with that of other systems.