Resilient distributed consensus for tree topology

Abstract

Distributed consensus protocols are an important class of distributed algorithms. Recently, an Adversarial Resilient Consensus Protocol (ARC-P) has been proposed which is capable to achieve consensus despite false information provided by a limited number of malicious nodes. In order to withstand false information, this algorithm requires a mesh-like topology, so that multiple alternative information flow paths exist. However, these assumptions are not always valid. For instance, in Smart Grid, an emerging distributed CPS, the node connectivity is expected to resemble the scale free network topology. Especially closer to the end customer, in home and building area networks, the connectivity graph resembles a tree structure. In this paper, we propose a Range-based Adversary Resilient Consensus Protocol (R.ARC-P). Three aspects distinguish R.ARC-P from its predecessor: This protocol operates on the tree topology, it distinguishes between trustworthiness of nodes in the immediate neighborhood, and it uses a valid value range in order to reduce the number of nodes considered as outliers. R.ARC-P is capable of reaching global consensus among all genuine nodes in the tree if assumptions about maximal number of malicious nodes in the neighborhood hold. In the case that this assumption is wrong, it is still possible to reach Strong Partial Consensus, i.e., consensus between leafs of at least two different parents.