Differentiated Services in Optical Burst-Switched Networks

Optical burst switching (OBS) [1] is a promising all-optical data transport technique to efficiently use the bandwidth offered by wavelength-division multiplexing (WDM) technology. An important issue in OBS networks is how to support quality of service (QoS). Despite the bandwidth availability, a link has at most few tens of wavelengths nowadays. Once, a burst occupies one wavelength, or a fraction of this, during the transmission some bursts will be blocked depending on the offered load to the network. In addition, the existing QoS mechanisms are proposed for packet switching networks and, at most, are based on management of electronic buffers [2]. To use these mechanisms in OBS networks, it is necessary to convert the optical signal to the electronic domain at each intermediate node, which limits the data transport rate. Furthermore, optical random access memories (RAMs) are not yet available. Bursts can be only delayed using fiber delay lines (FDLs) nowadays [3]. Thus, it is necessary to develop specific QoS mechanisms for OBS networks. Several mechanisms have been proposed for providing service differentiation in OBS networks [3], [4]. Zhang et al. [4] propose two admission control mechanisms: a static and a dynamic mechanism. Both are based on the number of wavelengths occupied by each service class. In the static mechanism, a fixed set of wavelengths Wi in a given link is reserved for bursts of a given service class i. In the dynamic mechanism, a fixed number of wavelengths Wi, not a fixed set, is reserved for bursts of a given service class i. In this two mechanisms, every node must keep track of the number of wavelengths occupied by bursts of each service class to guarantee that the number of wavelengths occupied by bursts of a given class i does not exceed Wi. As consequence, each node must store a great number of states. We propose an admission control mechanism for providing QoS in OBS networks. The proposed mechanism admits bursts of a given service class according to network load and a class-associated parameter. Based on this parameter, referred as load level, it is possible to differentiate the burst blocking probability experienced by each service class. We also develop an analytical model for the proposed mechanism. The performance of the three mechanisms static, dynamic and the proposed mechanism is analytically evaluated according to the blocking probability experienced by service classes. Different scenarios are tested by varying the offered load and the traffic amount of each service class. The results show that the proposed mechanism properly differentiates the services in all analyzed scenarios and always provides a lower blocking probability for the high-priority class in comparison with the other two admission control mechanisms.