Burstiness and resource management in ATM networks

Networks based upon the asynchronous transfer mode (ATM) provide for high flexibility to cope with a wide range of applications, some of them producing highly sporadic traffic. Therefore, the problem of burstiness has become in the last few years a key issue for such networks. A basic question is how to dimension network buffers in the presence of bursty traffic? In this paper we investigate the concept of burstiness and its impact on resource management. In burstiness characterization encountered in the literature, special attention has been given to the squared coefficient of variation of inter-arrival time Cv² in a cell arrival process. In order to observe the impact of bursty traffic on a queue, we develop in the present paper a ‘bursty’ traffic model, namely the two-stage hyper-Bernoulli cell arrival process, HBP₂, for short. We numerically solve the HBP₂/D/1/K queue. We especially derive the rejection probability P_loss. Numerical results are then thoroughly studied and we discuss the relevance for burstiness characterization of Cv² and peak to mean rate ratio B. We draw attention to the concept of local overload, i.e. when the arrival rate is greater than the server rate. This seems to be the most relevant phenomenon in the impact of a bursty traffic on a queue. These results are finally applied to the problem of resource management in ATM networks.