Studies on Fairness Improvement and Effective Wavelength Utilization in OBS Mesh Networks

Badr Oulad Nassar (0751144)

In optical burst switching (OBS) networks, an optical burst is generated from multiple IP packets coming from access networks. After the optical burst is generated, its corresponding control packet reserves all the wavelengths until the destination node, and then the optical burst is transmitted along the reserved wavelengths, all optically, hence providing an ultra fast transmission speed. In OBS networks, the number of wavelength reservations is equal to the number of hops, and hence, the higher the number of hops the higher the burst loss probability. Such an unfairness is one of the most important issues in OBS networks.

In order to resolve unfairness, hop-based burst-cluster transmission have been proposed and its performance was evaluated in a ring network. However, in mesh networks, two problems arise: first, the overall burst loss probability increases significantly due to synchronization; second, fairness is degraded because the amount of traffic on each link is not necessarily the same.

In this presentation, in order to resolve the synchronization issue, we propose the utilization of random scheduling for hop-based burst-cluster transmission. In random scheduling, the maximum waiting time is determined in advance from the maximum acceptable delay of an IP packet and the buffer size of the ingress node. Every time a burst-cluster is generated, a scheduler derives the actual waiting time at random, based on the obtained maximum waiting time, and then the burst cluster is transmitted from the scheduler.

Moreover, in order to adapt hop-based burst-cluster transmission when the traffic load is dynamic, we propose dynamic burst ordering. In this method, each ingress node calculates the burst loss probability for each number of hops using ACK and NACK messages. Based on the calculated probabilities, the ingress node changes the order of bursts within a burst-cluster dynamically.

The simulation results show that both proposed methods improve the performance of hop-based burst-cluster transmission significantly in terms of burst loss probability and fairness.