Paper detail

Computing backup forwarding rules in Software-Defined Networks

The past century of telecommunications has shown that failures in networks are prevalent. Although much has been done to prevent failures, network nodes and links are bound to fail eventually. Failure recovery processes are therefore needed. Failure recovery is mainly influenced by (1) detection of the failure, and (2) circumvention of the detected failure. However, especially in SDNs where controllers recompute network state reactively, this leads to high delays. Hence, next to primary rules, backup rules should be installed in the switches to quickly detour traffic once a failure occurs. In this work, we propose algorithms for computing an all-to-all primary and backup network forwarding configuration that is capable of circumventing link and node failures. Omitting the high delay invoked by controller recomputation through preconfiguration, our proposal's recovery delay is close to the detection time which is significantly below the 50 ms rule of thumb. After initial recovery, we recompute network configuration to guarantee protection from future failures. Our algorithms use packet-labeling to guarantee correct and shortest detour forwarding. The algorithms and labeling technique allow packets to return to the primary path and are able to discriminate between link and node failures. The computational complexity of our solution is comparable to that of all-to-all-shortest paths computations. Our experimental evaluation on both real and generated networks shows that network configuration complexity highly decreases compared to classic disjoint paths computations. Finally, we provide a proof-of-concept OpenFlow controller in which our proposed configuration is implemented, demonstrating that it readily can be applied in production networks.