AMPS: Application aware multipath flow routing using machine learning in SDN

This paper proposes an application-aware multipath flow routing framework that integrates Machine Learning Techniques (MLT) in Software Defined Networks (SDN). Applications generated by the devices are diverse in nature, for each application bandwidth and delay requirements vary. The flows in the network compete for a constrained resource such as bandwidth or low latency path, an intelligent flow routing algorithm becomes a natural demand. Better overall network performance could be achieved only if the network is capable of prioritizing the flows and assign resources based on their application specific requirement. Our proposed, AMPS controller is capable of prioritizing each of the flow using MLT, and assign a path based on its classified priority. AMPS controller supports routing flows through different path even if the flows are between the same pair of nodes. The path finding algorithm employs Yen-K-shortest path algorithm, also it supports scalable flow routing for a large volume of flows. We have implemented the flow routing algorithm in OpenvSwitch as a proof of concept. A significant improvement is observed in comparison to SDN with traditional routing techniques involving a large number of flows. The proposed flow routing algorithm ensures high availability of an unloaded path for high priority flows even in a heavily loaded network.