Cumulative sum techniques in ATM traffic management

Effective ATM traffic management requires knowledge of when traffic conditions change. This paper examines the use of cumulative sum techniques, widely used by the manufacturing industries, as a change point detection technique in ATM networks. The paper describes the one- and two-sided versions of the test, the use of average run lengths as performance measures and the determination of test parameters. The paper then examines scenarios where the test may be applied in ATM. Two example applications are chosen. One is to detect a change in Hurst parameter in long range dependency traffic and the other is a change in mean of a Markov modulated Poisson process. These applications illustrate the wide application areas of the test and address issues such as uncertainty in traffic parameter estimation and serial correlation.