A Novel Deep Q-Network-Based Scheme for Online Virtual Link Embedding in Software Defined Networks

With the advent of Software Defined Networking (SDN) and network virtualization, network systems are becoming increasingly more flexible and more scalable. With a global view and a centralized control, SDN can assign paths for individual virtual link (VL) requests, by employing classic graph algorithms (e.g. Dijkstra’s algorithm). However, this presumes that VL requests are independent of each other, which doesn’t hold true in real-world scenarios where VL requests can arrive randomly. In fact, previously instantiated VLs can take up resources (e.g. bandwidth) in the substrate network and render the residual resources unbalanced, leading to sub-optimal path assignment for subsequent VLs. This work investigates such scenarios and proposes a novel scheme based on Deep Q Network (DQN) and Link Feature Representation (LFR) for efficient online VL embedding. Experiments show that our proposed scheme (DQNLFR) can not only efficiently and effectively meet the routing requests of various VLs, but also ensure the balance of virtual network resources, leading to a higher acceptance rate. Moreover, the proposed solution has also the advantage of self-adaptability, scalability and generalization ability.