Bayesian Machine Learning Algorithm for Flow Prediction in SDN Switches

Software Defined Network (SDN) has been devised to solve many problems related to the current structure of the computer networks. SDN reallocates the control planes of all Networking Function Devices (NFDs) to a central entity (called controller) and keeps forwarding planes locally at each NFDs (called switches). Notwithstanding, segregation of data and control plans impose latency and overheads to SDN-based networks as a NFD needs to consult the controller how to handle each traffic. In order to overcome such shortcoming of SDN, this paper makes use of the Bayesian Machine Learning (BML) to allow switches to infer the underlying stochastic process by which controller classifies packets into flows. Based on this inference a switch can assign those packets whose flows are not given previously by the controller to the most appropriate flow. Extensive simulation conducted to assess the performance of the proposed algorithm highlights its advantages compared to the standard mechanism defined in the-state-of-the-art SDN implementation.