A Target Selection Scheme for Learning-Based Switch Migration in Distributed Software-Defined Networks

In distributed software-defined networking with multiple controllers, traffic variations can easily cause load imbalance among individual controllers. Thus, switch migration (SM) techniques have been introduced to address this problem. However, appropriate selection of the target controller for SM considering the dynamic nature of networks remains a challenge. In this paper, a learning-based SM (LSM) scheme is proposed to select the most appropriate target controller for SM operation. LSM employs the expectation-maximization algorithm to maximize the likelihood value of the potential target controller by learning the OpenFlow Packet-In message forwarding history. The experimental results demonstrate that LSM substantially outperforms existing schemes in terms of throughput and packet loss rate.