NI-HotStuff: A Reputation-Driven Committee Framework for Efficient and Robust BFT Consensus

Abstract

Consensus mechanisms are fundamental to blockchain systems, ensuring that distributed nodes agree on the validity of transactions and data. However, performance bottlenecks, particularly those related to throughput, latency, and node selection, have increasingly constrained the scalability of modern blockchain deployments. To address these issues, this paper proposes NI-HotStuff, a reputation-driven committee-based BFT consensus framework built upon the HotStuff protocol. A CatBoost-based reputation model is introduced to learn and evaluate historical behavioral features of nodes, enabling quantitative reputation scoring. A hardware-aware bidding mechanism is further incorporated to dynamically compute each node’s bid value and integrate it with its reputation score, thereby prioritizing stable and high-performance nodes for consensus participation. Moreover, a committee mechanism is established in which a set of  committee nodes were selected from the candidate pool, and only committee members participate in the consensus process, reducing redundant communication and mitigating the performance drag caused by weak nodes. On top of that, a leader-selection strategy based on reputation values and inter-view time intervals is designed to prevent low-reputation or potentially malicious nodes from frequently becoming leaders. Experimental results demonstrate that NI-HotStuff significantly outperforms traditional PBFT and HotStuff in terms of communication overhead, consensus latency, and system throughput, with particularly notable improvements in small- and medium-scale node environments.