DRL-based routing algorithm with guaranteed loss, latency and bandwidth in SDN networks: Application of online video conferencing

Abstract

Technology development in traffic engineering has led to the re-architecting of Online Video Conferencing Services (OVCS) through Software-Defined Networking (SDN) to provide scalability and low-latency performance. Since OVCS requires significant network bandwidth, routing algorithms must consider issues such as path redundancy, network congestion, and the rate of requested resources. With this motivation, this paper proposes a routing algorithm based on Deep Reinforcement Learning (DRL) to handle online video conference connection requests in SDN, which can simultaneously guarantee packet loss, latency, and bandwidth. The learning agent is configured based on the Asynchronous Advantage Actor-Critic (A3C) algorithm, which can adapt to the current traffic conditions and reduce the service latency as much as possible. We adjust the weights of the links in the weighted shortest path algorithm based on a set of critical nodes to improve the convergence ability of A3C and reduce the dependence of the learning agent on the network topology. Here, the link weight is measured by considering the critical nodes under maximum flow and minimum latency, configured to reduce interference in future routings. In addition, a postponement strategy is embedded in the SDN controller to prioritize requests based on the resources demanded for routing. This strategy is activated during network congestion and provides the ability to reserve more resources for future requests. The experimental results show that the proposed algorithm has a promising performance and convergence compared to the existing solutions and guarantees the QoS of OVCS. Specifically, the proposed algorithm can improve the throughput (up to 2.8%) and admission rate (up to 1.8%) of OVCS connection compared to the best existing solution.