| The two dominant approaches to parallel and distributed simulation, conservative and optimistic simulation, focus on implementing an event-scheduling simulation scheme. In this thesis we propose a new approach that is based on parallelizing an event-reservation simulation scheme. While in an event-scheduling approach only feasible events are scheduled and placed in a list of future events, in an event-reservation approach events that are only potentially feasible are placed in the future-events list, and in this sense potential future occurrence times of events are reserved. Event-reservation scheme, therefore, yields a more regular and less data dependent algorithm that is easier to parallelize. Moreover, in this scheme it is possible to use the same reserved events and times to simulate several--possibly many--structurally similar but parametrically different systems concurrently. The cost associated with using an event-reservation approach is due to generating null events, namely, potential events and event epochs that are not actually feasible and therefore not used. We give an analysis of different strategies that can be used for parallel and distributed implementation of event-reservation schemes and provide insight about key tradeoffs.; The Standard Clock (SC) simulation is one event-reservation based approach reported in the literature that is applicable to systems in which event lifetimes have bounded hazard rate. In this thesis, we introduce another event-reservation approach, called General Shared Clock (GSC) simulation, that removes the restriction on event lifetime distributions. This approach is based on some structural assumptions about the system being simulated and uses static event dependencies in the event set.; To illustrate the approaches introduced in this thesis and to demonstrate their effectiveness we apply them to the simulation of wireless communication networks (WCN), an area of great practical interest. We examine different implementation strategies and address related issues of system decomposition, synchronization, construction of correct sample paths, event utilization, inter-process communication, and efficiency. A detailed example of the simulation of a heterogeneous non-uniformly loaded WCN operating under different channel assignment policies--a challenging task both in terms of modeling complexity and computational intensity--is provided. We show that a locally-optimized distributed algorithm outperforms traditional centralized channel assignment schemes. |
