The trade-offs of QoS provisioning in IP-over-WDM networks with different switching timescales

Summary form only given. We consider centralised control as it allows us to implement a request scheduler residing in the control node and carry out QoS-based scheduling of the lightpath requests issued by newly aggregated bursts as well as the dynamic lightpath allocation to these bursts. Even though there is an overhead involved in the dynamic lightpath set-up in the centralised control, we demonstrate that the overhead becomes negligible if average burst size is maintained in the timescale of tens of ms. This way, the wavelength-routed optical burst switched network (WR-OBS) represents a fast circuit-switching scheme which is positioned in between wavelength-routed optical networks (WRON) and traditional OBS in terms of switching time. We show that the WR-OBS request scheduling can effectively provide low burst blocking probability (BBP) for both high and low priority traffic classes, whilst bounding end-to-end delays for each class. We also demonstrate that the desired QoS in WR-OBS can be achieved with higher bandwidth utilisation with respect to the traditional WRON. The trade-offs between BBP, end-to-end delays and wavelength requirement are presented.