Sensitivity analyses of parameters of a M(t) / G / [infinity] stochastic service system

Abstract

Parameter sensitivity analyses were conducted on a M(t) / G / [infinity] stochastic service system in which (1) the number of constants in an approximating nonhomogeneous Poisson process of inputs, (2) the mean of a Weibull c.d.f. of service time, and (3) the variance of the c.d.f. of service time were traded off in analyses of 24 cases for each of two fitting criteria: an L1 metric implemented by a linear goal program, and an L2 metric implemented by a multilinear least squares regression. The model goodness of fit and estimated total input to the system are both more sensitive to the mean service time than to its variance or to the number of constants in the approximating Poisson input. The fitting criteria give consistent results, but the L2 criterion gives slightly higher estimates of total input to the system over a fixed period of time.