AbstractOptical burst switching (OBS) appears as a promising technology for building dynamic optical transport networks. The main advantage of OBS is that it allows for dynamic allocation of resources at sub‐wavelength granularity. Nevertheless, the burst contention problem, which occurs frequently inside the network, has to be addressed before OBS can be really deployed as the next generation optical transport network. Recently a lot of attention is devoted to different approaches for resolving contentions in OBS networks. Although performance analysis of these approaches is strongly dependent on the traffic characteristics in the network, the majority of the studies is so far based on very hypothetical traffic assumptions. In this study we use traces of real measurements in the Internet to derive realistic data about the traffic that is injected into the OBS network. Specifically, we investigate the marginal distributions of burst size, burst interdeparture time, assembly delay and number of packets per burst as well as the burstiness of the burst traces. We demonstrate that the performance of an OBS core node using the real traces is pretty similar to the results obtained when the traffic arriving to the core node is assumed to be Poisson. In fact, usage of the Poisson as the process of burst arrival to the core node leads in all the investigated cases to an upper bound on the burst drop rate at that node. Copyright © 2009 John Wiley & Sons, Ltd.
Read full abstract