Analytic modeling methodology for evaluating the performance of distributed, multiple-computer systems (markov, petri nets)
A. Kapelnikov · 1986
In this dissertation, we describe an analytic modeling methodology for evaluating the performance of distributed, multiple-computer systems. The concepts and techniques of this methodology are useful for the approximate analysis of a wide range of distributed computing environments and communication networks. The main strategy of our approach is to segregate, as much as possible, the model of the logical behavior of an application (a program or a process) from the model of its underlying execution environment. For representing program behavior, graph-based techniques are used, while extended queueing networks are utilized for modeling system architectures. The solutions of both types of models are combined to estimate the performance of a distributed system in executing some selected applications. To illustrate the practical application of the methodology introduced in this dissertation and provide an indication of its expected accuracy level, we have included two case studies.