Link-based fine granularity flow migration in SDNs to reduce packet loss

Abstract

To maximize the data center network utilization, the SDN control plane needs to frequently update the data plane via flow migration as the network conditions change dynamically. Since each switch updates its flow table independently and asynchronously, the network state transition may result in serious link congestion and packet loss if it is done directly from the initial to the final stage. Deadlocks among flows and links may also block update processes. In this paper, we novelly migrate flows in a finer granularity of links, which is more likely to find a deadlock-free update plan. We prove that it is NP-hard to check the feasibility of a consistent flow migration. We also propose an efficient heuristic for allocating link resources to avoid deadlocks. We show the necessary and sufficient conditions for the deadlock existence in special situations. Extensive simulations show that our solution achieves a much higher probability of a consistent flow migration than prior methods.