Swarm Intelligence Based Dynamic Routing and Wavelength Assignment for Wavelength Constrained All-Optical Networks Ali Hassan * , Chris Phillips * and Zhiyuan Luo † * Department of Electronics Engineering Queen Mary, University of London, E1 4NS, London-UK E-mail: ali.hassan@elec.qmul.ac.uk † Dept. of Computer Science Royal Holloway University of London, TW20 0EX, Surrey-UK Abstract— This paper examines the problem of Dynamic Routing and Wavelength Assignment (DRWA) in Wavelength Division Multiplexed (WDM) networks with the wavelength continuity constraint applied. Traditional search algorithms like Integer Linear Programming and Graph Colouring become inefficient for solving optimization problems like DRWA because of the complexity of the task. Particle Swarm Optimization (PSO) inspired by swarm intelligence is proposed here to solve DRWA. A novel fitness function is designed for members of the swarm population that takes into account the normalized path length of the chosen route and the normalized number of free wavelengths available over the whole path. This normalization enables the approach to be self-tuning. Simulation results obtained for NSFNET and EON show that the proposed PSO- based scheme achieves a low blocking probability as compared to other swarm intelligence schemes like Genetic Algorithms (GA) for DRWA. Index Terms—Dynamic RWA, Swarm Intelligence, Wavelength Continuity, All-Optical WDM networks. I. INTRODUCTION Wavelength Division Multiplexed optical networks possess considerable bandwidth capacity [1] as separate optical channels can coexist on the same fibre link. Users communicate by optical channels called ‘lightpaths’ being setup between the provider-edge access points over a meshed network of links connected via Optical Cross Connect (OXC) devices. However, in order to route information along paths over these links generally requires a process call Optical- Electrical-Optical (O-E-O) conversion inside the OXC devices so that the information can be interrogated and re- modulated onto a different wavelength for each leg of its journey. In all-optical networks this process can be simplified allowing the information arriving at the OXC to remain encoded onto a particular wavelength that is simply redirected to the appropriate output, typically using Micro-Electro- Mechanical Systems (MEMS). The establishment of lightpaths creates logical circuits on top of the physical topology of the WDM optical network. If the intermediate nodes along the chosen route for the lightpath to be setup are not equipped with wavelength- conversion capability [2], then the same wavelength needs to be employed along its entire path. This restriction is known as the wavelength continuity constraint. Furthermore, two lightpaths sharing a common edge (i.e. link) of the network need to be assigned unique wavelengths as shown in Figure 1. In this case two lightpath share a common edge between node ‘7’ and node ‘10’, therefore each needs to be assigned a unique wavelength. This is called the “wavelength clash constraint”. 11 10 12 7 6 4 3 9 5 2 0 1 13 7 6 4 3 8 5 2 1 0 9 10 11 12 13 Lightpath (s) on Wavelength λ1 Lightpath (s) on Wavelength λ2 Optical Switch / Optical Cross connect Lightpath (s) on Wavelength λ1 Lightpath (s) on Wavelength λ2 Optical Switch / Optical Cross connect Figure 1: 14 Node, Wavelength Routed WDM NSFNET with Example Lightpath Connections When a connection request is made to setup a new end-to- end lightpath, a Routing and Wavelength Assignment (RWA) mechanism attempts to find an appropriate route and assigns a wavelength for it. The RWA approach can be categorized into two types: static and dynamic [3]. In Static RWA (SRWA), all the lightpath requests are known in advance, and the problem is to assign routes and wavelengths in a global manner, while minimizing the network resources [4]. In Dynamic RWA (DRWA), the lightpath requests arrive unexpectedly with random holding times. The objective of DRWA is then to find route and assign wavelength so that the blocking probability of the connection requests is reduced. SRWA is a well-known NP (Nondeterministic Polynomial- time) - Hard problem [5]. DRWA is more challenging because the lightpath requests arrive randomly and stay in the network for random amount of time. Traditional mathematical search schemes like Integer Linear Programming (ILP) and Graph Colouring become ineffective at solving the DRWA problem because of the computational complexity. Therefore, different heuristic and stochastic schemes are used to solve DRWA in a reduced amount of time. Most of the heuristic