Performance of Heterogeneous Parallel Computer System with Distributed Memory Using Analytic and Simulation Techniques

A heterogeneous parallel computer system with distributed memory can be described as a parallel computer system where each processor has its own queue and different departure rate. A good way of modeling the performance of the system is to consider the whole queuing network of the system, including the I/O queuing systems. Analytic queuing method of modeling performance has its limitation because of its inability to determine the exact convergence of some mathematical series that are involved in the modeling. Simulation queuing method that is based on recursive models can be used in performance modeling, but it has the limitation of exponential time complexity, especially when a recursive model has more than one recursive calls. This paper uses a combination of analytic queuing method and simulation queuing method to eliminate the above limitations.