[Paper Review] Iterative Approximate Consensus in the presence of Byzantine Link Failures

Lewis Tseng, Nitin H. Vaidya|arXiv (Cornell University)|Jan 26, 2014

Distributed systems and fault tolerance20 references4 citations

TL;DR

This paper presents a tight necessary and sufficient condition for solving iterative approximate consensus in arbitrary directed graphs under transient Byzantine link failures, where links may be compromised by an omniscient adversary. The method uses iterative averaging with bounded memory, leveraging matrix analysis and ergodicity bounds to ensure $ε$-agreement despite faulty links, proving convergence within finite, graph-dependent iterations.

ABSTRACT

This paper explores the problem of reaching approximate consensus in synchronous point-to-point networks, where each directed link of the underlying communication graph represents a communication channel between a pair of nodes. We adopt the transient Byzantine link failure model [15, 16], where an omniscient adversary controls a subset of the directed communication links, but the nodes are assumed to be fault-free. Recent work has addressed the problem of reaching approximate consen- sus in incomplete graphs with Byzantine nodes using a restricted class of iterative algorithms that maintain only a small amount of memory across iterations [22, 21, 23, 12]. However, to the best of our knowledge, we are the first to consider approximate consensus in the presence of Byzan- tine links. We extend our past work that provided exact characterization of graphs in which the iterative approximate consensus problem in the presence of Byzantine node failures is solvable [22, 21]. In particular, we prove a tight necessary and sufficient condition on the underlying com- munication graph for the existence of iterative approximate consensus algorithms under transient Byzantine link model. The condition answers (part of) the open problem stated in [16].

Motivation & Objective

• To address the problem of achieving approximate consensus in directed networks where communication links may be transiently compromised by Byzantine adversaries.
• To extend prior work on Byzantine node failures to the more general case of Byzantine link failures, which are increasingly prevalent in real-world networks.
• To identify a necessary and sufficient condition on the underlying communication graph for iterative approximate consensus to be achievable under the transient Byzantine link failure model.
• To design a lightweight, memory-efficient iterative algorithm that maintains validity and achieves $ε$-agreement without requiring global network knowledge.
• To resolve part of the open problem posed in [16] regarding the characterization of solvable graphs under this failure model.

Proposed method

• Models the network as a directed graph where each directed link represents a communication channel subject to transient Byzantine failures.
• Employs a restricted class of iterative algorithms that maintain only local state and do not require knowledge of the global topology.
• Uses row stochastic matrices to represent state transitions across iterations, ensuring validity through convex hull preservation.
• Applies matrix product analysis to bound the coefficient of ergodicity, proving convergence via $δ(\Pi_{u=1}^{t} \mathbf{M}[u]) \leq (1 - \beta^{rn})^{\lfloor t/rn \rfloor}$.
• Introduces a block-structured matrix product $\mathbf{Q}(i) = \prod_{t=(i-1)rn+1}^{irn} \mathbf{M}[t]$ to analyze long-term behavior and establish scrambling properties.
• Leverages the fact that $\lambda(\mathbf{Q}(i)) \leq 1 - \beta^{rn}$ to bound disagreement between nodes and ensure $\u03b5$-agreement within finite time $t_{\text{end}}$.

Experimental results

Research questions

• RQ1 What necessary and sufficient condition on the underlying communication graph enables iterative approximate consensus under transient Byzantine link failures?
• RQ2 Can iterative algorithms with bounded memory and no global topology knowledge achieve approximate consensus when links are subject to Byzantine corruption?
• RQ3 How does the presence of Byzantine links affect the convergence time and agreement bound in iterative consensus protocols?
• RQ4 To what extent do existing results on Byzantine node failures generalize to the more complex case of Byzantine link failures?
• RQ5 Is it possible to derive a tight, graph-theoretic characterization of solvability for iterative consensus under the transient Byzantine link model?

Key findings

• The paper establishes a tight necessary and sufficient condition for the existence of iterative approximate consensus algorithms under the transient Byzantine link failure model.
• The convergence time $t_{\text{end}}$ is finite and depends only on the graph structure, the failure bound $\beta$, and the desired $\u03b5$-agreement threshold.
• The coefficient of ergodicity $\delta(\Pi_{u=1}^{t} \mathbf{M}[u])$ decays exponentially as $\left(1 - \beta^{rn}\right)^{\lfloor t/rn \rfloor}$, ensuring rapid convergence.
• The algorithm guarantees $\u03b5$-agreement after $t_{\text{end}}$ iterations, with the final disagreement bounded by $|v_j[t_{\text{end}}] - v_k[t_{\text{end}}]| < \u03b5$.
• The proof relies on showing that each block matrix $\mathbf{Q}(i)$ is scrambling and has spectral gap $\lambda(\mathbf{Q}(i)) \leq 1 - \beta^{rn}$, ensuring contraction.
• The result resolves part of the open problem in [16] by providing a precise graph-theoretic characterization of solvability in the presence of Byzantine links.