A Coordinated Two-Stages Virtual Network Embedding Algorithm Based on Reinforcement Learning

Network virtualization allows multiple heterogeneous networks to be deployed on the same physical network. One of its main challenges is how to effectively allocate resources to the virtual network, which is known as the Virtual Network Embedding (VNE) problem. Due to the NP-hard characteristic of VNE, most of the existing approaches are based on the heuristic algorithms tended to converge to local optimal solutions. In this paper, we propose a VNE algorithm based on reinforcement learning (RL) methods. Firstly, the VNE problem is formulated as a combinatorial optimization problem. A pointer network including two Long Short-Term Memory modules is used to set up node mapping strategies. Then, an algorithm named active search based on policy gradient is designed to optimize the pointer network, and does not need the pre-training process and the search optimal policy for VN requests. The experimental results show that the proposed algorithm can improve the performance in average physical node utilization and long-term revenue to cost ratio.