DEVELOPMENT AND USE OF OPERATIONAL ANALYSIS MODEL ERROR MEASURES
Abstract
Assumptions that are the basis for operational analysis models of devices in networks have the characteristic that they can be warrented by observed data. This approach to calculating performance measures is not used in stochastic analysis which has some assumptions that cannot be proven by observing specific data. Error measures are defined for the main operational analysis assumptions. Terms are developed which are functions of these error measures and can be used to get exact results of particular behavior sequence performance measures, such as average queue length and response time, no matter how poorly the operational analysis assumptions are met. Limiting values of the error measures are found. Operational analysis formulas for performance measures that were developed assuming homogeneous arrivals and services are found to give exact results under less restrictive assumptions. Since the performance measure correction terms can only be calculated exactly with an amount of data that would be required to obtain direct performance measure results, ways to estimate the correction terms with reduced data collection are suggested. Two approaches are taken. One is to estimate the correction term based on partial data collection of a behavior sequence. The other approach is to determine bounds on performance measure correction terms, then to use these bounds to make statements about future correction term values. Comparisons of operational analysis models of average queue length and the stochastic models of the same performance measures are made. These models are shown to correspond for infinite capacity queues. A method for determining theoretical expected values of correction terms is given.
Degree
Ph.D.
Subject Area
Industrial engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.