Performance analysis of a cache-based multiprocessor system using closed queueing network model

An analysis for the performance measures (throughput and shared memory access time) of a cache-based multiprocessor system with multistage interconnection network is presented. The hit ratios of the shared data are obtained using Markov models based on various data movement policies. The developed closed queueing network model analyzes the performance of the multiprocessor system. Performance figures for 64*64 and 1024*1024 systems corresponding to these data movement policies are presented. The results show that the performance of such systems is dependent on the hit ratios of the data item requested. It is also indicated that the performance measures are dependent on the policy followed for the data movement in the entire system. The results show that keeping a single copy in the systems is better than the conventional global memory system.<
>