We consider the fundamental problem of updating arbitrary routes in a software-defined network in a (transiently) loop-free manner. Our objective is to compute fast network update schedules which minimize the number of interactions (i.e., rounds) between the controller and the network nodes. We first prove that this problem is difficult in general: The problem of deciding whether a k -round update schedule exists is NP-complete already for k=3 , and there are problem instances requiring Ω(n) rounds, where n is the network size. Given these negative results, we introduce an attractive, relaxed notion of loop-freedom. We show that relaxed loop-freedom admits for much shorter update schedules (up to a factor Ω(n) in the best case), and present a scheduling algorithm which requires at most Θ(logn) rounds.

Top