No abstract

No abstract

To answer the questions for the performance issues of multiple-bus multiprocessor systems under general conditions, a discrete-event simulation model has been developed. This model has been analyzed for different types of probability distributions for random variables A = processing times and B = memory access times for processors. For variables A and B, the following types of distributions are employed: exponential, Erlang, hypoexponential, hyperexponential, normal and constant. Simulation experiments show that the coefficient of variation of B, CB has a significant effect on the system performance, especially for CB>1, but the effect of CA is negligible if A is either hypo- or hyperexponentially distributed. However, this is not the case for a normal distribution. Processing efficiency, memory utilization and the percent of queued processors are some of the performance metrics used in our simulation model. We simulated several multiprocessor configurations and reach the conclusion that if the number of buses is equal to half the minimum value of the number of processors and the number of memory modules, the corresponding bus-deficient system behaves as a bus-sufficient system. Clearly, this is a substantial saving for the required number of buses without degrading the system performance.

A simulation model (program) is constructed for performance analysis of multiple-bus multiprocessor systems with shared memories. It is assumed that the service time of the common memory is either hypo-or hyperexponentially distributed. Processing efficiency is used as the performance index. To investigate the effects of different service time distributions on the system performance, comparative results are obtained for a large set of input parameters. The simulation results show that the error in approximating the memory access time by an exponentially distributed random variable is less than 6% if the coefficient of variation is less than 1, but it increases drastically with this factor if it is greater than 1.

Two different approximate state models are presented for the performance analysis of a sharedmemory multiprocessor system. The steady-state solutions for the state transition matrices are obtained by applying an iterative numerical technique. The numerical results illustrate that the state models produce less than 5% approximation error comparing to simulation results. It has been observed that the coefficient of variation for the service-time of the shared memory has a little effect on the system performance.