Markovian Queueing Network Models for Performance Evaluation of Multiple-Bus Multiprocessor Systems.

A closed-form solution for the performance analysis of multiple-bus multiprocessor systems is presented. A Markovian queueing network model has been developed to investigate the effects of memory and bus contentions on the multiprocessor system performance. The symmetrical structure of the Markov chains of the queueing network model makes it possible to demonstrate that local balance is satisfied. Consequently, the steady-state probabilities of the states of the Markov chains can be expressed by simple formulas. Processing efficiency is used as a primary performance measure and its relationship with the other performance measures are established. To investigate the effects of the system design parameters on the multiprocessor system performance, comparative results have been obtained for a large family of multiprocessor configurations from unibus to bus-sufficient systems. The simulation results show that if the st and ard deviation of the service time of the common memory does not differ too much from its mean value, then the error produced by the assumption of exponential distribution is in an acceptable range.Ph.D.Computer scienceUniversity of Michiganhttp://deepblue.lib.umich.edu/bitstream/2027.42/160576/1/8512481.pd