Analysis and improvement of traffic prediction in software defined networking

Abstract

Traffic prediction is an important part in the analysis and management of network. The prediction of network traffic provides on demand resource allocation and improves the quality of service. Network traffic can be predicted using statistical techniques as well as using artificial intelligence techniques. The statistical techniques are non-stationary in nature and may not be suitable for dynamic environments of 5G networks. The 5G networks are elastic in resources and dynamic in nature and provides most of the services using virtualization in on-demand manner. The artificial intelligence techniques like machine learning and deep learning are suitable for dynamic 5G environment. Traffic prediction using the machine learning algorithms like regression, support vector machine involves the identification of the network features by the network operator, but in deep learning models, automatic feature extraction from the traffic traces provides a more reliable solution without the interference of the network operator. The main objective of the current research is to design a lightweight traffic prediction model with minimal communication overhead. The spatial and temporal features of the network traces are captured efficiently for minimal prediction loss. The centralized learning model in Software Defined Networking controller increases the computational overhead as well as communication overhead in the controller. The distributed prediction model leverages the load between the controller and the switches, which reduces the computation overhead. The privacy of the network traces also is preserved since the local models in the OpenFlow switches exchange only the learning parameters with the centralized controller. newline