SDNProbe: Lightweight fault localization in the error-prone environment
Journal
Proceedings - International Conference on Distributed Computing Systems
Journal Volume
2018-July
Pages
489-499
Date Issued
2018
Author(s)
Abstract
Probe-based fault localization identifies potential faulty nodes, which are manually inspected for confirmation. This work explores efficient and accurate fault localization, which is crucial for reducing the manual effort without affecting network functionality. Prior work suffers from either high bandwidth overhead or false detection (i.e., incorrectly attributing good nodes or missing faulty nodes), especially in the presence of multiple or inconsistent faults. We propose SDNProbe, a lightweight SDN application that sends a provably minimized number of probe packets to pinpoint malfunctioning switches. We extend SDNProbe to randomize tested paths and packet headers to further improve the detection accuracy. Using realistic topologies and flow rules, our evaluation results confirm that SDNProbe can rapidly localize faulty switches while reducing the number of required test packets by 30%, compared to prior approaches. Even with 50% of switches being faulty, the extended SDNProbe can detect all faulty switches in 33 seconds, whereas prior approaches have false negative rates of 15-40%.
SDGs
Type
conference paper