Traffic load prediction in SDN/Open Flow Networks
KeywordsApplication software -- Development ; Future network traffic ; Software-defined networking environment ; Artificial neural networks ; OpenFlow SDN network
This thesis is an experimental attempt at applying machine learning techniques to predict the load of a network within an SDN (Software Defined Networking) environment. Utilizing an Autonomic Network Management Framework, which was implemented by the University of Piraeus, as well as actual network infrastructure with OpenFlow switches, an autonomous software component was developed using the Java programming language, which makes predictions of future traffic between five nodes on a network by implementing the Backpropagation algorithm. The software component is deployed and tested in the above Framework as an SDN application. The main objective of the conducted experiments was both to assess the efficiency of a well-known machine learning algorithm regarding prediction of future network traffic in real conditions based on specific network traffic patterns, as well as to investigate the possibility of simultaneous execution and cooperation of more than one SDN applications in order to facilitate the autonomic decision-making of traffic routing based on future predictions without any human intervention.