Algorithm design in optical networking

Abstract

Optical networks based on WDM technology provide cost-effective solutions for high-speed data transmission. Technological progress in optical networking has enabled WDM networks to evolve from traditional circuit switched ring networks to a new generation of packet switched mesh networks with high reliability. Nevertheless, WDM networks still face several challenges. First, it is impractical to implement OPS (Optical Packet Switching) as optical buffer technology has not yet reached maturity. Though a hybrid electrical/optical switch architecture can be designed by replacing optical buffers using their electrical counterparts, the reconfiguration overhead of optical fabrics generally leads to a large speedup requirement. OBS (Optical Burst Switching) removes the need for optical buffers and speedup, but it suffers from severe burst contention. Besides, burst delay due to offset time is quite large because of the O-E-O conversions of control packets at each intermediate node. Challenges also exist in optical network survivability design. Although monitoring cycle (m-cycle) has been introduced as an interesting technique for fast link failure detection in all-optical networks, we still lack an efficient algorithm for m-cycle design with minimum network resource consumption. Similarly, although the classical p-cycle (Preconfigured Protection Cycle) concept allows fast optical protection with high capacity efficiency, constructing a set of optimal p-cycles for 100% protection is still very time-consuming. In this thesis, efficient algorithms are designed to address these challenges. For optical packet switching, we aim at achieving 100% throughput with bounded packet delay in a hybrid switch architecture combining electrical buffers with an optical switch fabric. Four original traffic scheduling algorithms, ADAPT, SRF (Scheduling Residue First), αi-SCALE and QLEF (Quasi Largest-Entry-First), are proposed. ADAPT and SRF can automatically adapt themselves to switches with different parameters to minimize the required speedup. αi-SCALE and QLEF are minimum-delay scheduling algorithms. They always produce a schedule with the minimum number of switch configurations equal to the switch size. Based on the observation that OBS performance is significantly affected by the hop-count of the connections, LWMD (Least Weight Minimum Diameter) algorithm is designed to generate a virtual topology for an OBS network. This allows the control packets to be transmitted in a smaller number of hops, and thus reduces both the burst contention probability and the delay caused by offset time. For static routing in WDM networks, we propose an MET (Most-Even-Traffic distribution) algorithm to minimize both network wavelength requirement and network bandwidth requirement. We also propose a heuristic LTA (Light-Trail Assignment) for light-trail design. LTA can generate close-to-optimal solutions and is more scalable than the existing Integer Linear Programming (ILP) approach. We then focus on network survivability. For fast link failure detection in all-optical networks, an M2-CYCLE (minimum-length m-cycle) algorithm and an ILP-based approach are designed to construct m-cycles to minimize the network cost. The ILP-based approach allows non-simple m-cycles in the solution, and an efficient tradeoff between monitor cost and bandwidth cost. For p-cycle design, we formulate ILPs without candidate cycle enumeration. Four ILPs are formulated in this way, three for simple p-cycle design and one for joint simple and non-simple p -cycle design.