Classified Cloning for QoS Provisioning in OBS Networks Shavan Askar 1 , Georgios Zervas 1 , David K. Hunter 1 , Dimitra Simeonidou 1 1: School of Computer Science and Electronic Engineering, University of Essex, Colchester CO4 3SQ, UK, {skaske@essex.ac.uk} Abstract This paper introduces a novel cloning scheme for QoS provisioning in OBS networks. Results produced by the NS-2 simulator show 50% loss reduction in higher priority traffic when compared to the classical cloning scheme. Introduction Optical Burst Switching (OBS) combines the merits of Optical Circuit Switching and Optical Packet Switching while overcoming their shortcomings. It therefore appears to be an effective paradigm for efficient utilization of the huge bandwidth of wavelength division multiplexing to carry bursty IP traffic, particularly in metro networks 1 . There has been a rapid increase in the volume of traffic from new applications (such as video on demand, Voice over IP, online gaming or Grid computing) which have real-time and/or bandwidth constraints. Hence, service differentiation must be provided for such applications in order to reduce the loss rate while maintaining the lowest possible end- to-end delay. Most existing research in this area can be categorized into one of the mechanisms shown in figure 1. Contention resolution schemes appear to be a very tempting solution to the problem of contention in OBS networks, however there are a number of implementation problems; 1) wavelength conversion is an immature technique which is still very expensive to implement, 2) FDL’s are bulky and they merely offer fixed delays which will generally reduce channel utilization because they generate voids between scheduled bursts, 3) deflection routing suffers from the problem of endless loops as well as the possibility of insufficient offset time for rerouted bursts, 4) burst segmentation is still very complicated to implement. Reactive loss recovery is a retransmission scheme where burst retransmission is possible in the event of contention. Many factors hinder its implementation; firstly, very large buffers are required in ingress nodes in order to implement retransmission. Also, although retransmission may be practical in LAN’s, it is not useful in MAN’s or WAN’s because of their higher latency, which also requires larger buffers in order to implement retransmission. Finally, a notification protocol is required to notify edge nodes of burst losses, which generates additional load on the control channel. To overcome these problems we propose a new scheme for QoS provisioning with real time applications - Classified Cloning - which is inspired by the basic cloning scheme 3 . In this paper we investigate the use of cloning to reduce packet loss. Research conducted in the area is limited, with contradictions in the results from different studies 3,4 . However it has been shown that the existing drawbacks of burst retransmission, such as the large buffer size and increased control traffic, can be avoided through cloning, yielding lower mean packet delay. Basic Cloning Scheme In the classical Burst Cloning Scheme 3 (BCS) and our proposed Classified Cloning Scheme (CCS), the original copy of a burst is referred to as the “original burst”, and the duplicated copy as the “cloned burst”. Similarly, the traffic corresponding to the original and cloned bursts is referred as “original” and “cloned” traffic respectively. The node at which cloning is performed is referred to as the “cloning node”. In BCS, one or more cloned bursts can be made from each original burst and sent simultaneously; if one or more of these bursts arrive at the destination, the original burst is considered to be successful. On one hand, if more copies are made for a particular burst then it is less likely to be lost. On the other hand, if more copies are made overall, more cloned traffic is added to the network, which then Reliable OBS FDL WC FEC Loss Recovery Loss Minimization Proactive Reactive Retransmi ssion 1+1 Protection Composite BA Cloning Contention Avoidance Contention Resolution Admission control Deflection Segmenta tion Serializa tion Load Balancing Fig. 1. OBS Mechanisms for loss reduction ECOC 2010, 19-23 September, 2010, Torino, Italy 978-1-4244-8535-2/10/$26.00 ©2010 IEEE P5.04