Approaches for consistency-preserving simulation through correct detection of update dependency in distributed and parallel real-time simulation

Ignoring update-dependency relations among simulated items can lead simulation systems into inconsistent states. This becomes a particularly serious issue in distributed and parallel real-time simulation (DPRTS) requiring both high-performance simulation execution and real-time simulation support. An inadequate decomposition of simulation models for distributed execution can result in very high computational cost for detecting update dependencies among simulated items handled by distributed simulators executing on different nodes. In real-time simulation, all the simulation activities including such detection of update dependencies should be completed in a bounded time at every simulation step, and such a time bound can be a safe simulation step-size for all distributed simulators in DPRTS. Therefore, establishment of sound technical foundations for consistency-preserving simulation (CPS) through correct and timely detection of update dependency in DPRTS is an important task.; The update dependency and consistency properties are highly application-dependent. Thus, a domain-specific study of DPRTS of moving items in various dimensional spaces has been made in this dissertation to identify critical issues for realizing CPS in DPRTS. The identified issues are correct identification of all update dependency in the system, efficient state update of simulated items that are update-dependent upon each other, and derivation of a tight bound for the safe simulation step size.; The main contribution of this research work is a set of effective solutions for the above issues; specifically: (1) a fine-grain space-management method to minimize the search space for update dependency per simulated item, (2) a hierarchical territory-expansion technique to identify update dependency among the moving items under distributed simulation, (3) an ordered state-update method to avoid unnecessary re-simulation of simulated items, and (4) an iterative method to derive an analytical bound on the safe simulation step-size by analyzing three core activities of each simulation step. Moreover, two approaches, which incorporate the proposed solutions, have been formulated for simulating moving items in one-dimensional (1D), two-dimensional (2D), and three-dimensional (3D) space. In order to empirically evaluate their performance, both approaches have been implemented in a DPRTS testbed for simulating flying items in 3D space and the experimental results are discussed in this dissertation.; Fault-tolerance support is another important requirement for DPRTS. A typical design template for introducing real-time fault-tolerance support in an existing DPRTS system without any modifications to the original logic of the simulation system has been proposed through an exercise of rapid prototyping of a fault-tolerant DPRTS system. The real-time fault-tolerance capability of the prototype system has been empirically validated.