Solving Virtual Network Mapping Fast by Combining Neural Network and MCTS
Cong Wang, Xiaoyang Chen, Yixin Luo, Guoqi Zhang · 2022
Virtual network embedding (VNE) focuses on how to effectively allocate the limited substrate network resources to support more virtual network requests at the same time. Based on the combination of Monte-Carlo tree search (MCTS) and neural network in reinforcement learning, we propose an algorithm named VNE-NNMCTS to solve the VNE problem. The information including substrate network resources and virtual network requests is used as the state node in MCTS, the maximum confidence upper limit function (UCT) is used in MCTS to explore and utilize. The revenue to cost ratio is introduced as the reward to train the residual network to accelerate the global optimal search of the VNE problem. Compared with the typical traditional VNE algorithm, experiment results show that the proposed algorithm can improve the acceptance ratio, long term revenue to cost ratio and physical node utilization.