Bandwidth of prioritized crossbar systems

Abstract We analyze the performance of prioritized multiprocessor systems with shared memory using as interconnection networks the crossbar. The effective memory bandwidth (EMBW) is chosen as the main performance measure. One situation is examined: when the memory requests of each processor are of any distribution, and the rejected requests are resubmitted in the next cycle. A static model is derived for estimating the EMBW. Improvements are made by using an adjusted memory request rate. In order to verify these models, simulation results are given to compared with analysis results. In addition, we also give the extension of our analyses to include a nonhomogeneous crossbar system.