Congratulations to Manon Sourisseau, Jérémie Decouchant, François Taïani, and David Bromberg for their paper entitled “Partition Detection in Byzantine Networks” accepted at ICDCS 2024. Jérémie works at TU Delft; meanwhile, all remaining authors are members of the WIDE team.
The paper addresses the problem of detecting network partitions while considering the Byzantine fault model. In distributed systems, partitions arise when the network becomes divided into segments unable to communicate, often due to hardware failures, congestion, or software issues, potentially causing data loss and inconsistency. While partition detection in distributed systems has been extensively researched in the past decade, existing solutions, while resilient to crash faults, falter in the presence of arbitrary or malicious behaviors, commonly represented as Byzantine faults, where nodes can act in a manner detrimental to system correctness. The paper presents a novel algorithm called ‘Nectar’ that safely detects partitioned networks while tolerating Byzantine faults and proves its correctness.
An in-depth experimental evaluation shows that Nectar maintains a 100% accuracy while existing state-of-the-art baselines decrease by at least 40% as soon as one participant is Byzantine, all while maintaining reasonable bandwidth.
A preprint of the paper will be soon available.
