Efficient dynamic-committee BFT consensus based on HotStuff

Zongyang Zhang, Bin Hu, Lei Tian, You Zhou, Jianwei Liu

Abstract: Traditional Byzantine Fault Tolerant (BFT) consensus protocols are designed for fixed groups and are aimed to ensure the consistency of states among the replicas in distributed systems. Dynamic-committee BFT consensus protocols allow replicas to dynamically join and leave the system, enhancing the robustness and flexibility of distributed systems. However, the state-of-the-art dynamic BFT consensus protocol in partially synchronous networks suffers from an \(O(n^4)\) worst-case authenticator complexity, where n is the number of replicas. In comparison, existing static BFT protocols have achieved \(O(n^2)\) worst-case complexity. Hence, there is a performance gap between static and dynamic BFT consensus protocols. In this paper, we propose an efficient dynamic-committee BFT consensus protocol based on HotStuff, enabling member churn with minimal impact on performance metrics. With our improved committee reconfiguration technique, we reduce the worst-case authenticator complexity of dynamic BFT consensus from \(O(n^4)\) to \(O(n^3)\), while maintaining the best-case complexity of \(O(n^2)\). Besides, our protocol inherits the pipelined property from HotStuff, thus achieving a higher throughput. Experimental results show that our protocol has a peak throughput 4.2–7.6x as high as that of BFT-SMaRt, exhibiting a better scalability. The latency of join requests increases by 25%–60% compared to regular requests, while the latency of leave requests shows no significant difference from regular requests.