Design and Analysis of an Exponential Distribution Failure Detector for Fault-Tolerant Wired/wireless Networks

Failure detection is a fundamental issue for supporting reliability in wired and wireless fault-tolerant distributed networks, and is also an important performance bottleneck in providing generic service (similar to IP address lookup) in the event of node failure. It is very necessary to design an acceptable and optimized failure detector (FD) before the FD is actually implemented. Ensuring acceptable quality of service (QoS) is difficult due to the relative unpredictability of network environment. This paper proposes a novel Failure Detector, called Exponential Distribution Failure Detector (ED FD), which considers the probability distribution properties of arrival interval periods are exponential distribution, not normal distribution. In addition, we prove that ED FD belongs to the class ♦Pac (accruement property and upper bound property), and also analyzed the impact of history-track-record length on the performance of ED FD. Extensive experiments have been carried out over several kinds of networks (a cluster group, a wired LAN, a wireless network, and seven different WAN cases). The experimental results demonstrate that the impact of memory usage on the overall QoS is very small and can be negligible, and also demonstrated that the proposed ED FD outperforms the existing FDs in terms of short detection time, low mistake rate, high query accuracy probability, and good scalability.