Mining Historical Data towards Interference Management in Wireless SDNs.

WiFi networks often seek to reduce interference through network planning, macroscopic self-organization (e.g. channel switching) or network management. In this paper, we explore the use of historical data to automatically predict traffic bottlenecks and make rapid decisions in a wireless (WiFi-like) network on a smaller scale. This is now possible with software defined networks (SDN), whose controllers can have a global view of traffic flows in a network. Models such as classification trees can be used to quickly make decisions on how to manage network resources based on the quality needs, service level agreement or other criteria provided by a network administrator. The objective of this paper is to use data generated by simulation tools to see if such classification models can be developed and to evaluate their efficacy. For this purpose, extensive simulation data was collected and data mining techniques were then used to develop QoS prediction trees. Such trees can predict the maximum delay that results due to specific traffic situations with specific parameters. We evaluated these decision/classification trees by placing them in an SDN controller. OpenFlow cannot directly provide the necessary information for managing wireless networks so we used POX messenger to set up an agent on each AP for adjusting the network. Finally we explored the possibility of updating the tree using feedback that the controller receives from hosts. Our results show that such trees are effective and can be used to manage the network and decrease maximum packet delay.