An optimisation for a two-round good-case latency protocol

Abstract

Byzantine broadcast is a fundamental primitive in distributed computing. A highly efficient Byzantine broadcast protocol, motivated by the real-world performance of practical state machine replication protocols, is increasingly needed. This article focuses on the state-of-the-art partially synchronous Byzantine broadcast protocol proposed by Abraham et al. (PODC’21), which achieves optimal good-case latency of two rounds and optimal resilience of n ≥ 5f − 1 in this setting. Each step of the protocol is analysed, and then improved by cutting down the number of messages required to be collected and transmitted in the heaviest step of the protocol by about half, without adding any extra cost. This benefits from a new property, named “spread”, that we identify and extract from the original protocol. It helps us to eliminate non-essential work in its view-change procedure. The authors also show that no further reduction is possible without violating security. A prototype is implemented and the performances of improved and original protocols are evaluated in the same environment. The results show that our improvement can achieve about 50% lower communication cost and 40% shorter latency at a scale of 100 replicas. The latency gap becomes wider as the scale further increases.