Dissecting the Performance of Chained-BFT

TL;DR

This paper systematically evaluates chained-BFT protocols, introducing Bamboo as a benchmarking framework, and compares HotStuff, two-chain HotStuff, and Streamlet under various scenarios to reveal performance trade-offs.

Contribution

It presents Bamboo, a novel evaluation framework for cBFT protocols, and provides the first fair comparison of three major protocols with insights into their performance trade-offs.

Findings

HotStuff offers high responsiveness but lower forking-resilience.

Two-chain HotStuff balances responsiveness and resilience.

Streamlet emphasizes simplicity and robustness.

Abstract

Permissioned blockchains employ Byzantine fault-tolerant (BFT) state machine replication (SMR) to reach agreement on an ever-growing, linearly ordered log of transactions. A new paradigm, combined with decades of research in BFT SMR and blockchain (namely chained-BFT, or cBFT), has emerged for directly constructing blockchain protocols. Chained-BFT protocols have a unifying propose-vote scheme instead of multiple different voting phases with a set of voting and commit rules to guarantee safety and liveness. However, distinct voting and commit rules impose varying impacts on performance under different workloads, network conditions, and Byzantine attacks. Therefore, a fair comparison of the proposed protocols poses a challenge that has not yet been addressed by existing work. We fill this gap by studying a family of cBFT protocols with a two-pronged systematic approach. First, we present an evaluation framework, Bamboo, for quick prototyping of cBFT protocols and that includes helpful benchmarking facilities. To validate Bamboo, we introduce an analytic model using queuing theory which also offers a back-of-the-envelope guide for dissecting these protocols. We build multiple cBFT protocols using Bamboo and we are the first to fairly compare three representatives (i.e., HotStuff, two-chain HotStuff, and Streamlet). We evaluated these protocols under various parameters and scenarios, including two Byzantine attacks that have not been widely discussed in the literature. Our findings reveal interesting trade-offs (e.g., responsiveness vs. forking-resilience) between different cBFT protocols and their design choices, which provide developers and researchers with insights into the design and implementation of this protocol family.