Exploring Design Configurations of System Models: From Simultaneous Simulation to Search Heuristics

Simulation is applied in numerous and diverse elds, such as manufacturing systems, communications and protocol design, nancial and economic engineering, operations research, design of transportation networks and systems, and so forth. The real utility of simulation lies in the ability to compare and evaluate alternative designs before actual implementation or deployment of a system. To perform a thorough analysis of a large number of congurations with varying system design parameter values, it is important to develop efficient simulation and design space exploration methods that can evaluate a large number of alternative system congurations quickly and accurately. In situations where it is practical to exhaustively explore design parameter space, we proposed a new approach, called Simultaneous Simulation of Alternative System Congurations (SSASC), to evaluate dependability models that combines adaptive uniformization in simulation with the SCMS technique. SSASC showed that a signicant speed-up can be achieved compared to traditional discrete-event simulation to evaluate all alternative congurations. The event set management using adaptive clock algorithm and efficient data structures to manage system model's state access and update enables ecient simulation. Using SSASC, design engineers can benet from quicker evaluation of their system designs with better accuracy (due to variance reduction) than traditional simulation approaches provide. In situations where complete design exploration is not practical, this dissertation provides an intelligent search space exploration technique to effieciently determine near optimal solutions. This dissertation provides a technique, called design solver (DS), to determine near-optimal designs using meta-search heuristics. DS achieves efficiency in exploring the design parameter space by first determining the parameter values that have major impact on the quality of the design solution, and then determining parameter values that further fine-tunes the quality of the design solution. That decomposition reduces the size of the search space, allowing DS algorithm to focus on the most relevant regions to achieve a near-optimal solution. In essence, this dissertation "develops algorithms and techniques that would enable an ecient methodology to compare large numbers of alternative configurations in order to speed-up the design evaluation and validation process"