Intelligent Adaptive Routing Based on Multi-attribute QoS in Hybrid SDNs

With the emergence of new technologies and services, large-scale enterprise networks have strict and varied requirements for quality of service (QoS). In this paper, we design an intelligent routing switching scheme, named DIRH, to meet the diverse needs of multi-type traffic. In hybrid software defined networking (SDN) networks we use both centralized and distributed routing protocols coexist and communicate to various extents. Specifically, firstly, when a node is connected to a controller, we collect information such as delay, bandwidth, jitter, etc., and set weights to compute the link QoS. secondly, we use deep reinforcement learning (DRL) and graph neural networks (GNN) to capture the network topology and realize an efficient and adaptive routing policy. Then, when a node is disconnected from the controller, we implement distributed routing based on OLSR. The purpose is to guarantee the basic communication capability between nodes and reduce the delay caused by route discovery. We perform simulations using ONOS controller and Mininet. Extensive results show that DIRH achieves lower delay and jitter, as well as higher average throughput, compared to OLSR and DROM. As the number of nodes increases, the intelligence and effectiveness of routing can be better guaranteed.