Abstract-In the recent years, Software Defined Networking (SDN) has changed the way networks are engineered, making them more flexible, programmable and dynamic. SDN advocates for the centralization of control functionalities in a central node, the so-called controller. This entity has a wide view of the entire network, including the topology, facilitating the management and decreasing the complexity. However, the existence of a single entity running the complete control plane constitutes a single point of failure, thus triggering the need of improving the resiliency and reliability of the controller and the connection between the control and the data plane.This paper presents a solution for the improvement of the resiliency and the reliability on the OpenFlow channel through the use of multipath TCP (MPTCP). The proposed solution is based on the simultaneous use of in-band and out-of-band paths for the OpenFlow control channel, and includes a first experimental evaluation of the performance gains that can be achieved.Index Terms-SDN, OpenFlow, Resiliency, Reliability, MPTCP.A centralized controller is a potential single point of failure and a potential bottleneck. A failure on the controller or a disconnection between the control and data planes may lead to performance degradation and packet loss. OpenFlow uses a TCP connection (secured using SSL/TLS) to control the data plane elements. This connection, the so called OpenFlow channel, may use in-band or out-of-band transport signaling. In-band signaling is characterized by carrying the OpenFlow protocol packets through the same paths as the data transport, and therefore requires the network to be preconfigured to forward the OpenFlow signaling. Out-ofband signaling requires a separate network connecting all data plane elements to the controller, therefore requiring of extra network deployments.There is a need to improve the resiliency and reliability on software defined networks. Some efforts are devoted on the controller side, supporting a control plane distribution, starting with the version 1.2 of the OpenFlow protocol which supports mechanisms to use several simultaneous controllers. This concept was extended by diverse projects to increase the reliability on the control plane increasing the scalability and avoiding the single point of failure of a centralized architecture:• Onix [5] presents a platform for building a control plane on top of it as a distributed system. It provides an API which consists on a data model to abstract the network infrastructure, this API provides control logic, the possibility of read/write the state of the data plane and a notification engine for network state changes.• HyperFlow [6] solution is based on a logically centralized but physically distributed control plane synchronized with a publish/subscription system. All the controllers using HyperFlow have a consistent network-wide view and they run as if they are controlling the whole network.• ElastiCon [7] propose an elastic distributed controller architecture with a control...