An experimental investigation of dynamically reconfigurable computer network architectures through simulation

Abstract

The research described in this thesis is divided conveniently into three components: (1) the credibility assessment of a simulation model for the investigation of dynamically reconfigurable computer network architectures, (2) a comparative study of the standardized time series method of simulation output analysis, and (3) an experimental comparison of the effects of dynamic reconfigurability on message transmission delays and network throughput. The credibility assessment relies almost completely on verification procedures applied to both communicative and program representations of the model. In the absence of an extant system, validation consists of extensive, program traces to assure that model behavior matches expectations and reflects no inconsistencies. Application of a standardized time series technique produces the advantages reported by other researchers with regard to sampling efficiency (information derived per sample unit) when dynamic reconfigurability is precluded. The inherent non-stationarity induced by reconfiguration reveals the sensitivity of standardized time series and the consequent adjustment to preserve coverage. A compromise between coverage and sampling efficiency prompts the choice of the batch means method for experimental comparison. Experimental comparison shows that under high traffic variability reconfigurability