【文章內(nèi)容簡介】 nd analysis results. Finally, our conclusions and future works are discussed in Section 52 Related WorkRecent research results show that the routing in munication networks can be resolved efficiently by means of Ant Colony Optimization (ACO) [3]. The routing solution can be built using antbased agents behavior in their foraging of network states. These collective agents indirectly municate through pheromone trailing (stigmergy) in the environment. By following the pheromone trail of another, an agent can find a “good” route in terms of shortest, least congested path from the source to the destination to route the network data. Two basic algorithms are antbased control (ABC) for telephone networks, which was proposed by Schoonderwoerd et al. [4] and AntNet for packet switching networks, which was proposed by Di Caro et al. [5]. Some subsequent enhancement schemes to improve the antbased routing performance include smart agents which use dynamic programming [9], reinforcement learning which enhances the ant’s adaptability to its environment [10], and a genetic algorithm which adapts the ant control parameters to the search process [11]. While the Dynamic Routing and Wavelength Assignment in WDM Networks 831 above research focuses on the routing problem in electronic munication networks, our interest in this paper is the dynamic RWA problem in WDM optical networks with the constraint of wavelength continuity.Valera et al. [12] proposed an ACO approach for solving the static RWA problem. The goal is to minimize the number of wavelength requirement given a network topology and a traffic matrix. The wavelength assignment simply uses a greedy method that assigns the lowest available wavelength to each link. An ant’s route is selected based on the weight of attraction of each link. Each ant has its own pheromone that can be repulsed by others. Each ant keeps a “tabu” list of previously visited node for route backtracking and loop avoidance. The pheromone updating can use different methods。 the best result of this approach is obtained through global update when the weight of attraction of ant for a path increases with the number of traversed ants. The result can be pared to the conventional Nagatsu heuristic [13], but it requires a much longer putational time. However, this approach cannot be applied directly to the dynamic RWA problem.Garlick et al. [14] proposed an ACObased algorithm to solve the dynamic RWA problem. When a new connection request arrives, a number of ants are launched from the source to the destination. Ants evaluate a path based on its length and the mean available wavelengths along the path. Global pheromone updating is performed when an ant reaches its destination. The pheromone updating is on a perdemand basis: the network pheromone matrix is reset once a connection is established. The final best path for a connection request is made when all ants plete their exploitation tasks. The authors showed that this algorithm has better performance than an exhaustive search over all available wavelengths for the shortest path [15]. As a new set of ants must be launched for each new connection request, the setup delay will be very high due to the waiting for ants in large networks. In fact, this approach does not show the collective behavior of ants that e from different requests, which is an important aspect of antbased systems。3 AntBased RWA AlgorithmAn optical WDM network is represented by a graph with N nodes and E links. We assume that each link is bidirectional with a capacity of W wavelengths and no nodes have a wavelength conversion capability (wavelength continuity constraint). In order to support the route selection by ants, each network node has a routing table with N–1 entry. In a node i with ki neighbors, the routing table has a ki column. Each entry corresponds to a destination node and each column corresponds to a neighbor node. The value is used as the selection probability of neighbor node n when an ant is moving towards its destination node d. In order to support the wavelength assignment, we introduce the selection probability of each wavelength into the routing table. For each neighbor node, let be the probability that an ant selects the wavelength j when it moves to destination d.An example of the new routing table when W=2 is shown in . When a connection request occurs between source node 1 and destination node 6, node 3 will be selected as next hop because Wavelength 2 is preferred over wavelengt