Efficient techniques for improved QoS performance in WDM optical burst switched networks

Abstract

In this paper, we address the problems of burst scheduling, network fairness, and service differentiation in wavelength division multiplexing (WDM) optical burst switched (OBS) networks. We propose two approaches - time-slotting and burst fragmentation - to provide efficient solutions for the above problems. We also propose a path selection algorithm to distribute label switched paths (LSP) based on load balancing. Time slotting refers to the quantization of time into slots of fixed sizes. Time slotting aids fast implementation of scheduling algorithms. Further, the slotting is local to the nodes and they need not be synchronized across the network. In burst fragmentation, a burst is fragmented instead of being dropped if it cannot be accommodated on any wavelength as a whole. These fragments are then sent on different wavelengths on the same path. Only in the event, that a fragment cannot be scheduled on any wavelength of the link, is the burst dropped. We develop a new scheduling algorithm based on time-slotting and fragmentation. This algorithm has several attractive features such as fast and simple implementation and improved burst dropping performance. Using the hop-based shortest path algorithm to route a burst without accounting for the link load results in poor burst dropping performance. This is because, some links could be heavily loaded forming a bottleneck for network performance. Load balancing approach aims at reducing the bottleneck of data traffic in the network. We develop a path selection algorithm based on load balancing with the objectives of improving network performance in terms of burst dropping probability and improve fairness among bursts traversing paths of different lengths. We then develop an offset-based algorithm to provide inter-class service differentiation and intra-class fairness for the bursts belonging to classes with different levels of priority. An attractive feature of this algorithm is that it works with a constraint of maximum permissible initial offset time in order to reduce the burst transfer delay and also to reduce the buffer requirements at the ingress edge routers. It also ensures that shorter-hop bursts of a low-priority class does not perform better than the longer-hop bursts belonging to a high-priority class. Extensive simulation results have been used to demonstrate the effectiveness of the proposed approaches and algorithms for various identical and non-identical traffic demand scenarios. © 2004 Elsevier B.V. All rights reserved.