SDPredictNet-A Topology based SDN Neural Routing Framework with Traffic Prediction Analysis
Sowmya Sanagavarapu, Sashank Sridhar · 2021
Software Defined Networking is an intelligent network management approach for monitoring and improving performance such as in cloud computing. These networks follow separation between the Forwarding layer and the Control layer in the system to enable efficient programmability to perform network configurations. The SDN Controller present in the network has complete information about the network schema and its components to update the routing information of the switches present. The dynamic nature of traffic in these networks with multi-layer switches hinder the efficiency of SDN's performance in multi-cloud environments. This paper proposes SDPredictNet, a Recurrent Neural Network framework deployed on the SDN Controller that can predict the traffic in the network and update flow tables of the higher layer switches to perform routing based on the perceived bottlenecks in the network. SDPredictNet uses a Sequence-to-Sequence model that trains on the network data congestion to forecast the traffic in the SDN which is then modelled by an Artificial Neural Network to predict the path of the packets. SDPredictNet has achieved a RMSE score of 0.07 and an accuracy of 99.88% for traffic estimation and subsequent path determination.