On-the-fly Traffic Classification and Control with a Stateful SDN approach

The novel "stateful" approach in Software Defined Networking (SDN) provides programmable processing capabilities within the switches to reduce the interaction with the SDN controller and thus improve the scalability and the performance of the network. In our work we consider specifically the stateful extension of OpenFlow that was recently proposed, called OpenState, that allows to program simple state machines in almoststandard OpenFlow switches. We consider a reactive traffic control application that reacts to the traffic flows which are identified in real-time by a generic traffic classification engine. We devise an architecture in which an OpenState-enabled switch sends the minimum number of packets to the traffic classifier, in order to minimize the load on the classifier and improve the scalability of the approach. We design two stateful approaches to minimize the memory occupancy in the flow tables of the switches. Finally, we validate experimentally our solutions and estimate the required memory for the flow tables.