Enhancing TCP performance in IP fast reroute

IP fast reroute (IPFRR) mechanisms are efficient in providing protection against link or router failure by invoking locally determined backup paths. But in the presence of a transient network failure, the path oscillates between backup and original paths, causing frequent packet out-of-order arrivals at a TCP receiver. Duplicate acknowledgments (DUP_ACKs) generated by the receiver will then trigger unnecessary TCP congestion control at the sender. In this paper, we propose a novel yet simple algorithm called duplicate acknowledgement suppression (DAS) for enhancing the TCP performance in the presence of IPFRR. DAS only requires a minor modification to the TCP implementation at the receiver side while leaving the sender intact. The key idea is to use the time-to-live field of an out-of-order packet to infer the cause of its out-of-order arrival. If that is deemed due to a transient network failure, no DUP_ACK will be generated. Simulation results show that our DAS algorithm yields noticeable goodput improvement.