Performance of Compute Intensive Applications of Distributed Memory Heterogeneous Parallel Computer System Using Analytic and Recursive Models

Abstract— A distributed memory heterogeneous parallel computer system can be described as a heterogeneous parallel computer system where each of the processors has its own memory. In a heterogeneous parallel computer system, the computational resources distributed to each of the processors, like memory capacity and processor speed differ from one another. Since the memory is distributed, it means that each of the processors will have its own queue. Furthermore, a compute intensive application of a heterogeneous parallel computer is an application of heterogeneous parallel computer system where the overall utilization factor is high and greater than 1, for each of the heterogeneous queues of the parallel computer system. This means that the arrival rate into each of the heterogeneous queues is greater than the departure rate of the processor for that queue. On the other hand, a non-compute intensive application of a heterogeneous parallel computer system is an application of a heterogeneous parallel computer system where the overall utilization factor of each of the queues is less than 1. This paper models the various performance metrics of a compute intensive application of a distributed memory heterogeneous parallel computer system using analytic and recursive models.