Modeling burst traffic using a MAP with a tree structure

In this paper, a modeling approach for burst traffic using a MAP whose underlying process has a tree structure (TSMAP) is proposed. Characteristics of a MAP/G/1 queue are analytically obtained. For large batch size, however, there are elaborate calculations. The spectral method is applied to overcome this problem. The rate matrix of a TSMAP is estimated by the EM algorithm. For a trace-driven simulation and our analytical model, probability density functions of the number of data units in the queueing system are quite similar. Using short-term dynamics of queueing system, the long-term dynamics is estimated.