Learning and flow control in optimistic simulation

This thesis has two main contributions. The first contribution is the development of a modular, easy-to-use Time Warp simulation engine targeted towards distributed-memory environments. The second contribution is the analysis and experimental verification of the performance of the flow control algorithm proposed by Choe in a distributed-memory environment.The Time Warp simulation engine TWSIM provides our laboratory with a research medium for Time Warp simulations in a distributed-memory environment such as a network of workstations. The modular design of TWSIM allows for easy integration of any new simulation application and for fast testing of optimizations and improvements to the Time Warp mechanism. Its compact size and object-oriented implementation using the C++ programming language result in a short learning curve for future users and developers.The flow control algorithm proposed by Choe was implemented and analyzed with the aid of the TWSIM simulation engine. The algorithm makes use of stochastic learning automata to balance simulations loads by continuously regulating the flow of events between processors during the course of the simulation. Three different load metrics are considered: memory usage, virtual time, and a space-time product of the first two metrics. The algorithm was tested with two different simulation applications: a queuing network simulation and a Personal Communication Services (PCS) simulation. Results show that the flow control algorithm reduces the memory usage; the number of rollbacks and the number of antievents at the expense of the simulation time. As well, it becomes apparent that the behaviour of the flow control algorithm is not a consequence of learning.Finally, we discuss a number of approaches to learning and flow control using the outlines of the flow control algorithm, and we consider the extent of the performance improvement to be expected from memory-based schemes for limiting Time Warp optimism in a distributed-memory environment