Machine Learning Aided Load Balance Routing Scheme Considering Queue Utilization

Due to the rapid development of network techniques, packet-switched systems experience high-speed growth of traffic, which imposes a heavy and unbalanced burden on the routers. Hence, efficient routing schemes are required in order to achieve load balance. By decoupling the control plane and the data plane, Software-Defined Network (SDN) shows its flexibility and extensibility to achieve the automatic management of network resources. Based on the SDN architecture, we propose a pair of machine learning aided load balance routing schemes considering the queue utilization (QU), which divide the routing process into three steps, namely the dimension reduction, the QU prediction, as well as the load balance routing. To the best of our knowledge, it is the first time that principal component analysis (PCA) is used for the dimension reduction of the substrate network. Furthermore, QU prediction is conducted with the aid of neural network algorithms for the sake of coping with the network congestion resulting from burst traffic. Finally, simulation results show that our proposed routing schemes considering QU predicted by the machine learning algorithms outperform the traditional Bellman-Ford (BF) routing strategy in terms of the average packet loss ratio, the worst throughput and the average delay.