| Analyzing the functional behavior and performance of large concurrent systems is an important but difficult task. Often, the only feasible way to carry out the analysis is to simulate a model of the system, collect data from executions, and analyze the data. Conventionally, the simulation is carried out on one machine and for large systems it can take a long amount of time and require a large amount of memory. Distributed approaches to simulation can potentially overcome these limitations. The objective of this thesis is to suggest decentralized simulation techniques that are applicable to large problems.; The modeling approach used is to view a system as a set of interacting processes, where each process specifies the functional and timing behavior of an entity in the system. A distributed simulation of the system is then carried out by concurrently executing the processes on a multiple processor system. In this thesis, we are concerned with systems where the interactions between the entities can be modeled as occurring at discrete times.; It is well known that deadlocks may occur during distributed simulations which, if not resolved efficiently, cause a significant performance degradation. To overcome the problem, a distributed simulation algorithm is proposed that uses a hierarchical decentralized approach to localize overheads in resolving the deadlocks. Because the algorithm takes advantage of locality, it is particularly suited to simulation of systems with a large number of concurrently executing entities. The solution is generalized to simulations of systems with dynamic topology. Overheads associated with the algorithm are computed analytically so that appropriate hierarchical structures can be designed. An implementation of the system on a cluster of SUN workstations is discussed.; For monitoring and analysis of a concurrent system, a distributed mechanism called the Global Information Manager, based on the {dollar}langle{dollar}event, action{dollar}rangle{dollar} paradigm, is proposed. It collects and analyzes information from a distributed computation. The system can be used to record a trace of events occurring asynchronously in the system, analyzing the data from the events, and controlling the computation based on the analysis. |
