Analysis and design of optical burst switching networks

Optical Burst Switching (OBS) is a hybrid technique between coarse grain optical circuit switching and fine grain optical packet switching. In OBS networks, user data is switched entirely in the optical domain, while control and management functions are performed i n the electrical domain. This separation of the data plane and the control plane allows OBS networks to provide reasonably high levels of utilization while circumventing the need for optical buffering. In spite of OBS favorable features, several issues need to be addressed before OBS can be deployed in the Internet backbone. The objectives of this thesis are twofold: devise new methods for quality-of-service (QoS) provisioning in OBS networks, and develop new wavelength scheduling algorithms for enhancing the blocking probability in OBS networks. QoS provisioning is a major research problem in OBS networks. This is mainly because of the absence of the concept of "packet queues" in OBS networks. This thesis proposes two approaches for QoS provisioning in OBS networks. The first approach is a simple, yet effective scheme, called preemptive prioritized just enough time (PPJET). PPJET provides better service for high priority traffic by dropping reservations belonging to lower priority traffic using a new channel scheduling algorithm called preemptive latest available unused channel with void filling (PLAUC-VF) . Simulation results show that PPJET outperforms offset-based QoS schemes both in terms of dropping probability and end-to-end delay. As a second approach for solving the QoS problem in OBS networks, we present a detailed architecture for providing quantitative QoS guarantees with respect to endto- end delay, throughput, and packet loss probability in labeled OBS networks. The architecture describes a novel approach for applying fair scheduling algorithms in both the data plane of labeled OBS edge nodes and the control plane of core nodes without the need for optical buffering. In addition, we present analytical results for delay, throughput, and blocking probability in the proposed architecture. Simulation results demonstrate that the proposed architecture provides accurate and controllable service differentiation in labeled OBS networks. The absence of optical buffers in OBS nodes, coupled with the one way nature of OBS signaling protocols, drives the blocking probability to become the main performance measure in OBS networks. This give rise to the need for analytical models for calculating the blocking probability in OBS networks. In this thesis we present an approximate analytical model for calculating the blocking probability in OBS networks. The proposed analytical model takes into consideration the peculiar characteristics of OBS networks. To verify its accuracy, we compared the model results with results from a discrete-event simulation model. The proposed model results are in satisfactory agreement with simulation results. The blocking probability at an OBS node depends to a certain degree on how efficiently can the wavelength scheduling algorithm handle voids on wavelength channels. This fact has led to a growing interest in the area of wavelength scheduling in OBS networks. In this thesis we survey all previously proposed wavelength scheduling algorithms for OBS networks. In addition, we present two new wavelength scheduling algorithms: Min-AV and Max-NGV. We compared the performance of the newly proposed algorithms to those previously proposed using discrete-event simulation. Simulation results show that, in general, the Min-AV algorithm performs better than all previously proposed algorithms. Previously proposed wavelength schedulers, in addition to the Min-AV and Max-AV algorithms, are considered to be greedy algorithms. They are greedy in the sense that they consider every reservation request individually, and make the choice that looks best at the moment. We present in this thesis a new class of wavelength scheduling algorithms for OBS networks. The proposed wavelength scheduling algorithms process a batch of reservation requests together, instead of processing them one by one, and accept the requests that will maximize the utilization of the wavelength channels. We describe an optimal batch scheduler that serves as an upper bound on the performance of batch scheduling algorithms. Furthermore, we introduce four novel heuristic batch scheduling algorithms. Simulation results suggest that batch schedulers could significantly decrease the blocking probability in OBS networks.Applied Science, Faculty ofElectrical and Computer Engineering, Department ofGraduat