P4Resilience: Scalable Resilience for Multi-failure Recovery in SDN with Programmable Data Plane

Network failures are common in all kinds of networks, from data centers to wide-area networks. The current network mainly adopts reactive and proactive failure recovery schemes. The reactive scheme needs to contact to the controller to implement failure recovery after detecting a link failure, which will cause a large recovery delay. Whereas, the proactive scheme pre-calculates and installs backup paths in the switch to accelerate failure recovery, which consumes a lot of switch storage. Programmable data plane creates a possibility to solve network failure problem. To realize an efficient failure recovery while keeping switch storage overhead low, we present P4Resilience, a hybrid multi-failure recovery system combining SDN and programmable data plane. Firstly, P4Resilience encapsulates the backup path information into the packet header to perform failure recovery, which can greatly reduce the switch storage overhead. Secondly, P4Resilience implements loop-free backup path algorithm to calculate backup paths for each link to realize strong resilience to multi-link failure. Finally, we develop an optimization traffic balancing algorithm to reallocate bandwidth resources for flows to avoid potential congestion after failure recovery. Experimental results show that compared with the traditional failure recovery mechanisms, P4Resilience significantly reduces switch storage overhead for backup baths. Meanwhile, P4Resilience effectively relieves link congestion caused by failure recovery and guarantees the protection performance with a higher failure recovery rate and less packet loss rate.