PERFORMANCE EVALUATION OF HIERARCHICAL SIMULATORS: DISTRIBUTED MODEL TRANSFORMATIONS AND MAPPINGS (DISTRIBUTED SIMULATION, ANALYTIC MODELLING, DISCRETE EVENT

The objective of this dissertation is to develop a coherent methodology for performance evaluation of hierarchical simulators by integrating existing approaches and new results. For this objective, we have applied Zeigler's theory of discrete event specification systems (DEVS) to provide a framework for performance modelling and simulation. Our theoretical work has been checked experimentally with models of hierarchical distributed simulators expressed within one of the conventional discrete event simulation languages, SIMSCRIPT.;The performance evaluation methodology consists of converting a distributed simulation model to a composition tree representation, and then transforming and mapping the model onto a hierarchical simulator model. The resulting hierarchical simulator model is simulated within an appropriate experimental frame in order to evaluate its performance. Using DEVS formalism, we built such a hierarchically specified abstract performance simulators, called the Hierarchical Multiport Simulator (HMS), in modular form by coupling together simulators for each of the component DEVS. Each processor of the hierarchical simulator has several ports to communicate with other processors by message passing.;To evaluate the performance of hierarchical simulators, the analytic and simulation modelling approaches are employed. For analytic modelling, activity networks, called the delay graphs, are introduced, and critical path analysis is done to estimate simulation execution times in the delay graphs. For simulation modelling, the objective-driven methodology and modularity concepts are used in implementing the performance simulation program. By the use of the both approaches, optimal configurations can be found in terms of performance criteria such as speed-up or throughput.;Finally, we propose a technique to design and implement simulation systems to evaluate the performance based on experimental aspects. Experimental runs were made to determine the effect of the variations: the number of levels and branches, the type of transformation, the number of destinations and influences, the values of the delay parameters, and the ratio of the number of external tasks. These parameters were found to be major factors in affecting the performance of the hierarchical simulators. Our experiments indicate that simulation approaches are suited to find the optimal configurations of a given set of distributed simulation models.