Measurement-estimation approach to efficiency evaluation of bandwidth allocation scheme in ATM networks

It is essential for the development of bandwidth allocation schemes to study the resource management and traffic control in ATM networks. It is important to accurately evaluate the efficiency of various bandwidth allocation schemes. However, in the multimedia environment of ATM networks, it has been recognized that the performance analyses are very complicated because the source modeling is the non-Poisson arrival process. In this paper we propose a measurement-estimation based approach to analyzing the polling systems with server time-outs, and with Markov modulated Poisson arrival process (MMPP). These models can be used to estimate the efficiency of the DTS-like bandwidth allocation schemes. The idea is to transform the aggregated source stream into Poisson flow with constant rate during the fixed time interval. The method is to count cells in consecutive fixed intervals. The aggregated source state is estimated from the measured number of cells during the fixed interval. If the state is estimated to be j, the arrival process is assumed to be a Poisson process with rate j/spl lambda/ during the fixed interval. Therefore, the analytical results of the time-limited polling models with conventional simple source modeling can be used to evaluate efficiency of the bandwidth allocation scheme in ATM networks.