Traffic characterization for broadband networks

Traffic characterization is a difficult but critically important aspect in design, control and per formance analysis of broadband integrated networks. Selection of the traffic modeling approach should depend not only on the type and characteristics of sources, but also on the network architec ture (e.g., switching and multiplexing schemes), the level of the model (e.g., call-level, burst-level, or cell-level), quality of service (e.g., call blocking, cell loss, message delay). One must also take into consideration whether the network service to be modeled is loss-based or delay-based. In this presentation we shall review recent advances in modeling and characterization methods for multi-media traffic that is carried by high-speed networks such as ATM B-ISDN and AONs (All Optical Network). We first discuss probabilistic models of bursty traffic source(s): (1) the MMRP (Markov modulated rate process) model, which includes the well studied "on-off " source model; (2) the MMPP (Markov modulated Poisson process) model, which is designed to capture serial dependency in a superposed process of multiple sources; (3) diffusion process approximation models, which allow us to treat general sources without invoking computational complexity problems. We also review (4) the TES model, a novel method developed by Melamed et al. to synthesize traffic with a few parameters that can be used to drive a simulator. Finally we will discuss some outstanding problems in the area of traffic characterization that await further theoretical and empirical investigations.