Reinforcement Learning for Resource Mapping in 5G Network Slicing

In the era of 5G, network slicing is considered to be an effective solution for flexible network deployment and diversified services. It based on network virtualization technology to divide into multiple end-to-end virtual networks on the substrate physical network and in the form of services to meet the user that has customized appeal to network resource. This paper proposes a reinforcement learning algorithm based on collaborative relationship between node mapping and link mapping(RLCO). In the node mapping stage, we apply the policy network to calculate the probability of the physical node mapping to a virtual node. In the link mapping stage, we apply the Dijkstra algorithm. After the success of node and link mapping, the algorithm evaluates the mapping results respectively, and on this basis, defines the reward function of the RLCO algorithm. The difference characteristics of node and link are introduced into the reward function. The RLCO algorithm based on this reward function can make the result of network slicing mapping reach the global optimal. Furthermore, we compare the RLCO algorithm with the other three algorithms. The results show that the RLCO algorithm is superior to other three algorithms in terms of the acceptance rate of network slice requests and the long-term earning/cost ratio.