Control traffic balancing in software defined networks

Abstract

To promise on-line and adaptive traffic engineering in software defined networks (SDNs), the control messages, e.g., the first packet of every new flow and network traffic statistics, should be forwarded from software defined switches to the controller(s) in a fast and robust manner. As many signaling events and control plane operations are required in SDNs, they could easily generate a significant amount of control traffic that must be addressed together with the data traffic. However, the usage of in-band control channel imposes a great challenge into timely and reliable transmissions of control traffic, while out-band control is usually cost-prohibitive. To counter this, in this paper, the control traffic balancing problem is first formulated as a nonlinear optimization framework with an objective to find the optimal control traffic forwarding paths for each switch in such a way the average control traffic delay in the whole network is minimized. This problem is extremely critical in SDNs because the timely delivery of control traffic initiated by Openflow switches directly impacts the effectiveness of the routing strategies. Specifically, the fundamental mathematical structures of the formulated nonlinear problem and solution set are provided and accordingly, an efficient algorithm, called polynomial-time approximation algorithm (PTAA), is proposed to yield the fast convergence to a near optimal solution by employing the alternating direction method of multipliers (ADMM). Furthermore, the optimal controller placement problem in in-band mode is examined, which aims to find the optimal switch location where the controller can be collocated by minimizing the control message delay. While it is not widely researched except quantitative or heuristic results, a simple and efficient algorithm is proposed to guarantee the optimum placement with regards of traffic statistics. Simulation results confirm that the proposed PTAA achieves considerable delay reduction, greatly facilitating controller's traffic engineering in large-scale SDNs.