A Constructive Particle Swarm Optimizer for Virtual Network Embedding

Virtual network embedding (VNE) is a key technique for flexible sharing of physical resources in the modern Internet. As the VNE problem is nondeterministic polynomial hard, natural inspired population-based algorithms have been increasingly considered as promising approaches. But as the VNE problem has two tightly related tasks: node mapping and link mapping, it remains challenging for metaheuristics to deal with both tasks coordinately and effectively. This paper proposes a constructive particle swarm optimizer for virtual network embedding (CPSO-VNE). CPSO-VNE uses discrete vectors and matrices with probabilities to encode positions and velocities on VNE search space, respectively. What makes CPSO-VNE different from previous works is the step-by-step solution construction scheme introduced in CPSO-VNE. In this scheme, each node is mapped along with the mapping of its adjacent virtual links. In this way, node and link mappings are coordinated in one stage. In addition, with this step-by-step construction scheme, path information of networks can be introduced as heuristic information to guide the search, which can further improve performance. The proposed method is tested on the scenarios with a single VN request and with a set of online VN requests. The simulation results show that the proposed approach is promising.