Analysis of ATM statistical multiplexers and its application to connection admission control

Abstract

ATM networks require highly elaborated traffic control mechanisms. Connection admission control (CAS)is of great importance to achieve high resource utilization and to prevent network congestion. One essential problem in dealing with CAC issues is the performance evaluation of ATM statistical multiplexers, whose input traffic is the superposition of various traffic sources with different traffic characteristics and quality of service (QOS) requirements. Superposed ATM cell streams exhibit burstiness and strong autocorrelation properties. In this dissertation. we first propose a Markov modulated Poisson process (MMPP) modeling method to model the superposed ATM traffic. The proposed method enables us to identify two-state MMPP, MMPP(2), parameters from the distribution and autocorrelation functions of the cell interarrival times in traffic streams. For the interarrival times, the distribution and autocorrelation functions are modeled by a second-order hyper-exponential function and an exponential function, respectively. We calculate the cell loss ratio (CLR) of MMPP(2)/D/1/K system using the matrix geometric, and then compare the results with those of simulations as well as other modeling methods for individual source traffic modeled as the interrupted Poisson process (IPP). The proposed MMPP(2) modeling method shows good results especially in terms of the asymptotic decay rate for the CLR curves. For the superposed traffic in ATM networks, simulation show that the coefficient of variation of the interarrival times various in a wide range. For example, it is observed to be less than 1 for the superposition of several voice sources, nd greater than 1 for the superposition of several voice sources, and greater than 1 for the superposition of video sources with autoregressive moving average (ARMA) model. We propose a 3-state MAP, MAP(3), where the interarrival times have the second-order Coxian distribution. For thr superposed traffic, the Coxian distribution can...