A CONGESTION AND MEMORY-AWARE FAILURE RECOVERY IN SD-WAN

Abstract

In software-defined wide area networks (SD-WANs), link failure is a common occurrence creating heavy congestion and packet loss and degrading the application performance. Proactive failure recovery, i.e., preinstalling backup tunnels beforehand, is heavily used in SD-WAN to break the dependency on the controller and to enable fast rerouting. However, existing systems either lead to wasting the valuable network resources (such as bandwidth capacity and switch memory) because of vacant link capacity or impose non-realistic assumptions on the network topologies, such as the existence of link-disjoint routes or unlimited switch memory resources. We advocate a balanced network resource utilization instead. We argue that a reliable system must take into account multiple resources while planing a failure recovery and must be adaptable to multi-QoS traffic classes. Thus, in this thesis, we propose a novel multi-resource aware proactive recovery system in SD-WANs. Our system can be used in any network topology and is adaptable to environments with multiple QoS requirements. Moreover, it makes an optimized trade-off among the critical network resources and minimizes route stretch while recovering from a failure. In particular, we formulate the failure recovery problem as a multi-objective MILP optimization problem and evaluate it using CPLEX. Then we develop a heuristic to efficiently compute backup routes as the problem is NP-Hard. We implement a prototype of our system using the Ryu SDN controller and extensively evaluate it in Mininet over four real WAN topologies. The results show significant performance improvement of our failure recovery system compare to the state-of-the-art.