Hybrid modeling/simulation methodology for simulation speedup and preserved accuracy

Abstract

Recently discrete event modeling and simulation for performance evaluation of complex systems becomes an important research issue in many areas of system design. The main research objective is to devise a framework for developing accurate performance models and efficient simulation algorithms for fast experimentations with such models. As a part of such research activities, this thesis presents a hybrid modeling/simulation framework within which both accuracy in models and speed in simulation experimentations are obtained. We employ the Zeigler``s DEVS formalism, which supports hierarchical modular descriptions of discrete event systems. Though the set-theoretic formalism has expressive power and the well known simulator algorithm, it lacks of analytic means for reasoning about system behavior. To complement this shortage, we propose a model transformation algorithm which transforms selected DEVS models into behaviorally equivalent analytic ones in steady state as far as accuracy is preserved. For high-speed simulation, we develop a hybrid simulation algorithm which is capable of simulating both analytic and simulation models in a combined manner. To build an environment implementing such an algorithm, we extend the DEVSim++ environment by adding new classes for specifying and simulating analytic models. We call it HDEVSim++. HDEVSim++ is a result of the combination of two powerful frameworks for system simulation : the DEVS formalism and the proposed hybrid modeling and simulation methodology. Thus, it allows modelers to specify discrete event models using the hierarchical composition methodology in an object-oriented framework. HDEVSim++ also supports automatic model transformation and fast discrete event simulation. An advantage of our approach is that it employs a single modeling formalism, making the modeler possible to develop models using an expressive formalism. In addition, the developed framework is based on the sound theoretical foundation of behavio...