JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 27, NO. 15, AUGUST 1, 2009 3075 Traffic Engineering in the Wavelength Domain for Optical Burst-Switched Networks João Pedro, Student Member, IEEE, Paulo Monteiro, Member, IEEE, and João Pires, Member, IEEE Abstract—This paper addresses the possibility of using traffic engineering in the wavelength domain as a way to improve the performance of optical burst-switched (OBS) networks. With that aim, we detail a heuristic traffic engineering strategy that determines the order by which the ingress nodes of the network should search for an available wavelength for burst transmis- sion, in order to minimize the probability that data bursts going through overlapping paths reach a common fiber link using the same wavelength. By means of network simulation, it is shown that the proposed traffic engineering strategy outperforms the existing strategies in reducing burst loss in OBS networks. This trend holds for core nodes with different degrees of wavelength conversion, ranging from the absence of wavelength converters to the use of dedicated full-range wavelength converters. More specifically, it is shown that using only traffic engineering in the wavelength domain can dramatically reduce burst loss in networks without wavelength conversion or alternatively it can reduce the number of wavelength converters in networks based on shared wavelength conversion. The simulation results also show that by combining traffic engineering in the wavelength domain with the use of the in- expensive electronic buffers at the ingress nodes, the performance of OBS networks with dedicated full-range wavelength converters can be further improved at the expense of a slight increase in the burst transfer delay. Index Terms—Burst scheduling, optical burst switching (OBS), traffic engineering, wavelength assignment. I. INTRODUCTION T HE optical burst switching paradigm [1], [2] has at- tracted considerable interest as an optical networking architecture for efficiently supporting Internet protocol (IP) packet traffic, while exploiting the immense transmission ca- pacity available from optical fibers and the wavelength division multiplexing (WDM) technology [3]. OBS is envisioned as an intermediate evolutionary step between the already feasible optical circuit switching (OCS) and the conceptually ideal optical packet switching (OPS) paradigms. Manuscript received October 17, 2008; revised February 13, 2009. This work was supported by Nokia Siemens Networks S.A., Portugal. First version pub- lished May 02, 2009; current version published July 22, 2009. J. Pedro is with the Research, Technology, and Platforms Department, Nokia Siemens Networks S.A., 2720-093 Amadora, Portugal, and also with the Insti- tuto de Telecomunicações, Instituto Superior Técnico, 1049-001 Lisboa, Por- tugal (e-mail: joao.pedro@nsn.com). P. Monteiro is with the Research, Technology, and Platforms Department, Nokia Siemens Networks S.A., 2720-093 Amadora, Portugal and also with the Instituto de Telecomunicações, Universidade de Aveiro, 3810-193 Aveiro, Por- tugal (e-mail: paulo.1.monteiro@nsn.com). J. Pires is with Instituto de Telecomunicações, Instituto Superior Técnico, 1049-001 Lisboa, Portugal, (e-mail: jpires@lx.it.pt). Digital Object Identifier 10.1109/JLT.2009.2022042 Optical transport networks based on OCS can already be im- plemented with available optical technology. However, they are expected to be inefficient in supporting IP packet traffic mainly due to their coarse wavelength granularity, which can lead to significant overprovision of network resources. On the other hand, an OPS network provides statistical multiplexing at the packet level, improving the network resource utilization effi- ciency, but at the expense of advanced optical processing re- quirements, which are still too immature to be deployed. The OBS paradigm has brought the prospect of an all-optical network able to switch data with sub-wavelength granularity and using only moderately complex optical technology. At the ingress nodes of an OBS network, multiple IP packets directed to the same egress node are assembled into bursts, whereas at the core network wavelengths are allocated to each data burst during the time required for its transmission. Consequently, network re- sources are allocated to data with finer granularity and fluctua- tions in the traffic pattern can be more rapidly and accurately accommodated, when compared to that of an OCS network. Conversely, OBS technological requirements are less stringent than those of OPS because, similarly to OCS, the bandwidth for data burst transmission is reserved in advance by using out-of- band signaling, avoiding the need of complex optical processing devices. OBS networks use one-way resource reservation for setting up the necessary resources for transmitting each burst [4]. As a result, instead of waiting for an acknowledgment of successful resource reservation in the entire burst path, the transmission of a data burst is initiated soon after it has been assembled at the ingress edge node. This OBS feature implies that different bursts may contend for the same resources at a core node of the network. Unresolved resource contention is critical since it leads to burst loss, which degrades the network performance. Resource contention at the core nodes of an OBS network can be resolved using strategies acting in one or several of three domains: wavelength, time, and space. Contending bursts can be converted to other wavelengths using wavelength con- verters, delayed using fiber delay lines (FDL), or deflected to other output fiber links of the node. Wavelength conversion is the preferred contention resolution strategy [5], because it efficiently resolves contention and does not cause additional IP packet latency. Additionally, its effectiveness is independent of the network topology and traffic pattern. Hence, most OBS studies and proposals assume contention resolution based on full-range wavelength converters, which can convert any input wavelength to any output wavelength. However, all-optical wavelength conversion devices are still at the research and development stage [6], and remain expensive, complex and 0733-8724/$26.00 © 2009 IEEE