Response Surface Analysis of Stochastic Network Performance

摘要

The objective of this thesis was to analyze stochastic binary networks for the purpose of improving their performance as measured by expected maximum flow and source-to-sink reliability. The capacity and survivability of the networks' nodes and arcs formed the parameters of interest in the experimental design used to develop a response surface model. Estimates of network performance was provided by Monte Carlo simulation using a FORTRAN based program designed for this study called MAXFLO. MAXFLO implemented an original form of maximum flow calculation using minimal cuts instead of paths to improve the simulation's speed. MAXFLO was also compiled and run on a VAX 8650, VAX 11/785, and SUN-3 workstation under UNIX and VMS systems to insure portability and simulation performance. Keywords: Monte Carlo method, Network flows, Statistical analysis, Experimental data. (kr)