Byzantine Fault Tolerance in Distributed Constraint Optimization Problems

Koji Noshiro, Koji Hasebe

Keywords: Distributed Constraint Optimization, Byzantine Fault Tolerance, Replication

Abstract: Fault tolerance is crucial for the reliability of multi-agent system applications. This paper addresses fault-tolerant algorithms for solving distributed constraint optimization problems (DCOPs) in the presence of Byzantine faults, the most critical type of fault. First, we define a novel class of DCOPs to handle faulty agents, namely a Fault-Tolerant DCOP (FT-DCOP). This class extends the standard DCOP formulation to explicitly introduce the concept of faulty agents. Additionally, we prove an impossibility result for FT-DCOPs regarding the allowable proportion of faulty agents, revealing fundamental limitations. Furthermore, we propose a DCOP algorithm for solving FT-DCOPs that combines the well-known Max-Sum algorithm with replication. Unlike naive replication, which relies on complex Byzantine consensus protocols and strong assumptions, the proposed algorithm employs a simple verification step based on a synchronous message-passing mechanism and the deterministic nature of Max-Sum, thereby avoiding Byzantine consensus and relaxing assumptions. Finally, we experimentally evaluate the performance of the standard Max-Sum and the proposed algorithm in the presence of Byzantine faulty agents. The results demonstrate that the standard Max-Sum algorithm is vulnerable to faulty agents and generates low-quality solutions, while the proposed algorithm is resilient and matches the fault-free performance of Max-Sum with manageable communication overhead.

Area: Search, Optimization, Planning, and Scheduling (SOPS)

Generative A I: I acknowledge that I have read and will follow this policy.

Submission Number: 619