⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯ This work was partially funded by the Spanish Ministry of Science and Education under contract FEDER-TIC2002-04344-C02-02. Impact of Burst Length Differentiation on QoS performance in OBS networks M. Klinkowski, D. Careglio, S. Spadaro, J. Solé-Pareta Advanced Broadband Communications Centre (CCABA) Universitat Politècnica de Catalunya (UPC), Jordi Girona, 1-3, 08034 Barcelona, Catalunya, Spain Email: {mklinkow, careglio, pareta}@ac.upc.edu, spadaro@tsc.upc.edu ABSTRACT In this paper we address the problem of Quality of Service (QoS) provisioning in OBS networks focusing in efficiently support of two different classes of service, namely High Priority (HP) – to transport loss/delay sensitive traffic like real-time voice; and Best Effort (BE) – for regular data traffic. First, we provide a brief overview of the state of the art on the addressed problem focusing on the main drawbacks of presented both Offset Time Differentiation and Burst Pre-emption (BP) schemes. Then we describe a Burst Length Differentiation (BLD) technique that aims to improve the performance when applied together with other QoS techniques. The performance results show that BLD bring profits in particular when combining with BP scheme. Keywords: Optical Burst Switching, QoS provisioning, Offset-Time Differentiation, Burst Pre-emption. 1. INTRODUCTION This paper addresses the problem of providing QoS in Optical Burst Switching (OBS) networks. OBS is one of the potential solutions for future optical networks. On one side, it efficiently uses statistical multiplexing in the optical layer to overcome the wavelength switching inefficiency. On the other side, it uses large data units called bursts to overcome the fast switching requirements of the optical packet switching (OPS) currently only available at the laboratory. In contrast to OPS, in OBS networks the client packets are aggregated and assembled at the edge nodes of the network into bursts (large variable length data units). Meanwhile, the control information is signalled out-of-band and transmitted prior to the burst in such a way that the intermediate nodes have sufficient time to reconfigure the switching matrix. Several works have been proposed in the literature dealing with burst aggregation, signalling issues, resource reservation and QoS provisioning. See for example [1-4]. In this paper we only focus on the QoS problem analysing the drawbacks of the current proposals. Therefore, we propose the Burst Length Differentiation (BLD) technique, which aims to improve the performance when applied together with other QoS techniques. The rest of the paper is structures as follows. Section 2 presents brief state of the art on QoS in OBS. In section 3 we define the BLD technique. In section 4, we evaluate the performance of the BLD scheme when is combined with two of the most relevant techniques used to provide QoS in OBS networks. Finally, in section 5 we conclude the paper. 2. STATE OF THE ART IN QOS ISSUES FOR OBS QoS provisioning in OBS networks is usually based on the DiffServ approach, which is mainly applied in three different ways: 1) exploiting an extra offset-time (OT) in so called Offset Time Differentiation (OTD) scheme, 2) assigning priority to the higher traffic class and 3) varying the burst aggregation parameters [1]. In the first case, an extra OT is assigned to high priority bursts what favors them during resource reservation process in the network. In the second, QoS is provided in the core switch by preempting low priority bursts (with or without segmentation) or rescheduling low priority control packets. Finally, varying the burst aggregation parameters can help in minimization the end-to-end delay of high-priority packets. The pre-emption introduces high signaling overhead and complexity, while in OTD we observe a sensitivity of the high priority class to burst length characteristics and extended pre-transmission delay that may not be tolerated by some real-time applications. A segmentation technique, which first was introduced in order to decrease burst loss probability in OBS networks and then to assist the QoS [9], also introduces high complexity. Therefore, the problem is still opened. We address this problem by combining the BLD technique, which is described in the following Section, with both the Offset-Time Differentiation (OTD) and the Burst Pre-emption.