An integrated framework for parallel simulation

Parallel discrete event simulation offers a significant speedup over the traditional sequential event list algorithm. This dissertation attempts to compare, contrast, and integrate the conservative and optimistic classes of protocols commonly used for parallel simulation. We first address the problem of transparent execution of a simulation model using three different conservative protocols namely, null message, conditional event, and a new protocol that is a combination of the two. Language constructs to express lookahead and dynamic communication topology are discussed. Performance evaluation on four types of benchmarks--queueing network simulation, synthetic benchmarks, an existing switch-level VLSI simulator, and existing gate-level VLSI simulators, is presented. We present a study of the issues of non-determinism, simultaneous events, lookahead, and their interplay in different conservative and optimistic simulation protocols. Analytical bounds are derived for the best and worst case performance of optimistic and conservative protocols. We identify simulation scenarios in which the optimistic and conservative protocols may outperform each other arbitrarily. For many systems, it may be the case that different subsystems possess contradictory characteristics such that whereas some sub-systems may be simulated efficiently using a conservative protocol, others may be more amenable to optimistic methods. We propose the design of a parallel simulation protocol, called ADAPT, that allows different parts of a system to be simulated using different protocols, monitors the parameters that may alter the suitability of a protocol to a subsystem, and dynamically alters the protocol accordingly.