Improving link failure detection and response in IEEE 802.11 wireless ad hoc networks

Wireless multihop ad hoc networks face a multitude of challenging problems including highly dynamic multihop topologies, lossy and noisy communication channels, and sporadic connectivity which contribute to frequent link failures. Rapid and accurate link failure detection is therefore important to maintain correct and optimum operation of network routing protocols. In this paper, we propose a unified link failure detection and recovery architecture (ulfra) which uses link layer feedback for rapid failure detection and packet salvaging for packet recovery. While link layer feedback and packet salvaging have been studied in simulations and simple experiments, no thorough experimental study have been undertaken to evaluate their real-world performance. This paper essentially fills this void as we implement and evaluate ulfra in an IEEE 802.11 multihop ad hoc network. Our experimental results show that link layer feedback, as modeled in current network simulators, actually performs worse than hello beaconing as it generates excessive false failure detections. To improve its performance, we implement a veto mechanism to reduce spurious detections. Experimental results show that the veto mechanism dramatically improves the performance of link layer feedback in terms of packet delivery, delay, and routing overhead as it considerably reduces the number of false detections. Compared with hello, it delivers 15-20% more packets at high node failure and 12-20% more at high network traffic.