A Task-Oriented Hybrid Routing Approach based on Deep Deterministic Policy Gradient

Abstract

With the development of communication and transmission technologies, the applications of Internet of Things (IoT) and telemedicine are increasingly sensitive to network latency. To meet this urgent demand, learning-based routing approaches emerge, with better performance and higher flexibility. These routing approaches can be divided into single-path and multipath, which have lower transmission delay and better load balancing respectively. However, the traffic scheduling policy is not only affected by the network states but also needs to consider the requirements for traffic tasks. We propose a Task-Oriented Hybrid Routing Approach (TOHRA) based on Deep Deterministic Policy Gradient (DDPG) in Software-Defined Networking (SDN). First, the hybrid routing optimization algorithm comprehensively considers the network states and task requirements, and outputs a task-oriented hybrid routing that combines the advantages of single-path and multipath. Moreover, we design a hop-by-hop traffic segmentation model based on DDPG to output the traffic segmentation ratio on hybrid routing to adapt to the changing transmission path. Experimental results show that TOHRA achieves the best performance of load balancing and network throughput. Especially in the case of Type C tasks in Germany topology, the average network throughput of THORA is increased by 32.86%, and the average variance of link load rate is reduced by 46.62%.