QoE Oriented Cognitive Network Based on Machine Learning and SDN

Abstract

With the aim of improving application Quality of Experience (QoE), this paper presents a framework for a QoE oriented cognitive network that enables the implementation of a machine learning model in SDN architecture. Software Defined Networking (SDN) technology is applied to dynamically manage and orchestrate end-to-end network resources as per application needs and network condition and scenario. A structured approach is applied to implementing Machine Learning (ML) techniques within the network. A ML approach is intended to be used to autonomously learn the best management strategy for a given application and best fulfill its requirements. The framework is based on the combined SDN and ML approach, combining information obtained from both the SDN North Bound Interface (NBI) and South Bound Interface to assess both the network and application state and condition. A module built on the SDN controller uses this information to correlate network level metrics with application condition. This module will learn how the features of the network effect the application condition. This information is then used to make decisions with regards to network resources for the application. The framework is structured into three main modules: data collection and aggregation, network/application learning and network management with prediction. The proposed framework is intended to be used for investigation into link types and their effects on end-to-end path selection for application flows in SDN. This is essential for future networks as more diverse applications are expected to enter the mobile domain with each application flow traversing a range of link types.