Comparing Evolutionary Approaches for Routing in Dynamic Optical DWDM Networks Michael Taynnan Barros † , Paulo Ribeiro L. Júnior ‡ , Rafael Fernandes Lopes § and Marcelo S. Alencar ‡ Institute for Advanced Studies in Communications (Iecom) † Systems and Computing Department - DSC ‡ Electrical Engineering Department - DEE Federal University of Campina Grande (UFCG), Campina Grande, Brazil § Federal Institute of Education, Science and Technology of Maranhão - IFMA Email: {michel.taob,paulo,rafaelf,malencar}@iecom.org.br Abstract—The minimization of blocking probability is an important problem for the design of next generation high speed networks. This problem is usually approached with an adaptive routing algorithm and traffic grooming, that is costly regarding the necessary improvement on network equipment. To advance the solution of this problem two evolutionary approaches, the clonal selection and the genetic algorithm, are presented and compared. The blocking probability is used for comparison, as well as, the route size and the routers utilization, with the GÉANT2 topology and dense traffic. I. I NTRODUCTION Almost 5,6 billion users and 50 billion entities will be active in the Internet, in 2020, according to estimates [1]. This increase in the number of Internet users leads to intense research in smart technologies regarding quality of service (QoS) provision for the next generation networks (NGNs). This requires low transmission delay, large bandwidth, high availability and low blocking probability, for high data traffic. The use of dynamic optical networks is an option to build the needed infrastructure and to focus on techniques or technolo- gies to provide more connections, as new routing algorithms and multiplexing techniques. Dense wavelength division multiplexing (DWDM) is the premiere transport technology for long-haul and metro or regional networks [2]. It greatly expands the network capac- ity over the existing network infrastructure by simultaneous transmission of hundreds of wavelengths on a single fiber [3]. A unique feature of optical DWDM networks is the tight coupling between routing and wavelength selection. Therefore, to establish an optical connection, one must deal with both routing (selecting a suitable path) and wavelength assignment (allocating an available wavelength for the connection) to obtain a lightpath [4]. The problem is referred to as the routing and wavelength assignment (RWA) [5] and the performance of a RWA solution is measured based on the blocking probability (percentage of blocked connections). RWA is important for the control plane of DWDM networks and has received attention from the researchers of optical com- munications. Several RWA algorithms have been developed for dynamic routing, with the routes computed in real time, based in the state of the network. Recent advances in this subject are found in [6], [7] and [2]. Two interesting approaches for RWA in dynamic DWDM optical networks, called Optical Genetic Adaptive Routing Algorithm (OGARA) and Clonal Selection Adaptive Routing Algorithm (CSA) are proposed to minimize the blocking probability. They are based on evolutionary optimization al- gorithms for IP networks [8]. The main goal of the new algorithms is to achieve a performance similar or above the adaptive least used routing algorithm with traffic grooming, without the need to adjust the hardware or the use of wave- length converters [9]. The obtained results show a performance gain of 35%, comparing the CSA and the OGARA algorithms. Other approaches, based on adaptation of optimization algorithms, have been presented in the literature. Sinclair [10] presented a hybrid algorithm, which involved genetic algorithm (GA), but the genetic operators do not adapt well to the network condition and restrictions. A joint routing and dimensioning technique that used GA was presented in [11], but it does not attack directly the RWA problem. The remaining of the paper is organized as follows. Sec- tion II presents a formulation of routing and wavelength assignment. Section III presents the proposed OGARA al- gorithm. Section IV presents the proposed CSA algorithm. Section V presents the simulation environment and Section VI shows the analysis of the results. Section VII summarizes the paper. II. THE ROUTING AND WAVELENGTH ASSIGNMENT FORMULATION The physical topology of an optical network corresponds to the cabling installing and the devices. This structure can be represented by a graph G =(V,K), which has a set V =(v 1 ,v 2 , ··· ,v |V | ) of vertices (nodes) and a set K = (k 1 ,k 2 , ··· ,k |K| ) of edges (links), and each edge connects a pair of nodes in V . The graph G has a list of edges R, R ⊂ K, which represents the shortest path between a source node v s and a destination node v d . In optical networks, R is valid if the wavelength continuity constraint is considered, as in the following.