Browse Theses

Modeling real-world large-scale systems is inherently challenging due to their size and complexity. Detailed simulation of large-scale models requires tremendous amounts of computing power and memory space. Parallel discrete-event simulation techniques can overcome the limitations of a sequential simulator. The ultimate challenge for parallel simulation is to provide an efficient tool to model real-world problems and, at the same time, hide the complexity of parallelization as much as possible. In this thesis, we report our research advances in improving conservative parallel simulation techniques to deal with large-scale models. Our major contribution has been the investigation of novel conservative parallel synchronization algorithms that can significantly reduce overhead and thereby improve the performance of executing large-scale models on parallel computers. These algorithms include a lock-free asynchronous algorithm for scheduling logical processes, a composite synchronization protocol that can adapt to a wide range of model topologies, and a two-level synchronization scheme for enabling parallel simulation in a distributed-memory environment. These techniques are embedded in a high-performance parallel simulator, called Dartmouth Scalable Simulation Framework (DaSSF), designed to significantly simplify model development and increase model expressiveness and maintainability. We emphasize performance and memory efficiency by carefully examining key design features, such as process-orientation, and their associated overheads. Experiments show that DaSSF achieves excellent performance in a detailed simulation of large-scale networks on parallel architectures. We also apply parallel simulation techniques in large-scale wireless ad-hoc network models. We explore the use of a high-performance model to accelerate the simulation of the IEEE 802.11 wireless LAN protocol. And we improve the lookahead property, by exploiting characteristics of the wireless models, to reduce the synchronization overhead and achieve superior performance.