Learning-based wavelength assignment in optical burst switching networks

Abstract

As in any WDM network, the wavelength conversion capability of routers has a major effect on the performance of OBS networks. However, wavelength conversion is an immature and expensive technology and still remains as a topic of further study. Therefore assuming full wavelength convertibility is not a practical assumption. It is important to focus on avoiding burst drops in OBS networks with wavelength conversion incapable nodes, and efficient wavelength assignment is an important issue for this purpose. Wavelength assignment techniques proposed for Wavelength-Routed Networks are generally not suitable for OBS Networks and some distributed assignment techniques are needed to be devised. In this paper, we proposed a learning based wavelength assignment technique, namely LWA. We make some extensions to this technique via employing a preemption mechanism and using some features of OBS like burst aggregation, and show that the proposed techniques are suitable in order to reduce the burst drop probability, as well as supporting QoS and reducing the end-to-end delay.