PIQoS: A Programmable and Intelligent QoS Framework

Network management and Quality of Service (QoS) support are becoming more challenging with the increase in network traffic, size, and service requirements. To meet these challenges, we need a programmable and intelligent framework for automated QoS support; static or threshold-based approaches are not adequate. We propose a software-defined and machine-learning-based intelligent QoS framework called PIQoS. PIQoS enables software-defined networking (SDN) controllers to effectively, efficiently, and autonomously react, in a vendor agnostic way, to changes in network links by (1) placing link failure detection and recovery in the data plane and (2) applying supervised learning methods to the tasks of dynamically detecting link failures and congestion and appropriately reconfiguring the network so that it can continue to provide the required QoS as link properties change over time. To test the performance of a system based on PIQoS, we performed extensive simulation experiments in Mininet, using real network topologies. We also studied the comparative performance of several supervised learning methods applied to our specific detection and correction problems to determine which methods are most appropriate for this domain. Our simulation results highlight the potential efficacy of the PIQoS framework when applied in real networks.