Dynamic control for failure recovery and flow reconfiguration in SDN

Abstract

By separating control and data planes, Software-Defined Networks (SDN) pave the way to a plethora of online routing mechanisms that react instantaneously to sudden traffic variation and network failures. In this paper, we introduce a dynamic routing system where these online mechanisms strive to solve an optimization instance that constantly evolves due to new arrivals and departures as well as network failures and repairs. We model the optimality gap, which is characteristic to the surcharge cost paid by network operators, of such a system as an auto-regressive stochastic process and we show that its evolving average does not diverge. We further propose a control policy that operates on top of the online routing solver to dynamically decide when to reconfigure the network. The policy minimizes the average surcharge cost and guarantees that the network reconfiguration rate does not exceed a threshold set by the operator. Numerical results show that our policy outperforms periodic schemes and approaches that reconfigure constantly the network after any system changes occur.