Reputation-Based Streamlet: An Enhanced BFT Consensus Algorithm for Improved Performance and Scalability

Bo Fang, Dancheng Zhao, Taotao Wang, Hao Wang, Shengli Zhang, Qing Yang

Abstract: This work presents Reputation-Based Streamlet (RBStreamlet), an improved version of the Streamlet consensus algorithm, designed to address limitations such as high bandwidth load and high communication complexity during the Propose and Vote phases. RBStreamlet introduces a reputation mechanism for leader election and fork selection, along with erasure coding and Merkle trees to reduce data transmission volume in the Propose phase. Furthermore, RBStreamlet leverages threshold signature technology to decrease the message complexity in the Vote phase. We implement the proposed RBStreamlet algorithm in a prototype blockchain system and build a network emulator to evaluate its performance under conditions resembling a practical wide-area network. Through extensive experiments, RBStreamlet demonstrates superior performance, scalability, and resilience to Byzantine attacks compared to the original Streamlet, especially in bandwidth-limited scenarios. These findings suggest its potential for practical applications in blockchain systems.