Towards Optimal and Practical Asynchronous Byzantine Fault Tolerant Protocols

Abstract

With recent advancements in blockchain technology, people expect Byzantine fault tolerant (BFT) protocols to be deployed more frequently in wide-area networks (WAN) as opposed to conventional in-house settings. Asynchronous BFT protocols, which do not rely on any form of timing assumption, are arguably robust in such a setting. Asynchronous BFT protocols have been studied since the 1980s, but these asynchronous BFT works mainly focus on understanding the theoretical limits and possibilities. Until the recent asynchronous BFT protocol, HoneyBadgerBFT (HBBFT), was proposed, the field received renewed attention.

Dumbo family, a series of our works on the asynchronous BFT protocols, significantly pushed those protocols towards practice. First, all complexity metrics are pushed down to asymptotically optimal, simultaneously. Second, we identify the bottleneck in the state of the art and revisit the design methodology, identifying and utilizing the right components, and optimizing the protocol structure in various ways. Last but not least, we also open the box and optimize the critical components themselves. The resulting protocols are indeed significantly more performant, the latest protocol can have 100K tps and a few seconds of latency at a reasonable scale. This thesis focuses on the latest three members of the Dumbo family. To begin, we solved an open problem by proposing an optimal Multi-valued validated asynchronous Byzantine agreement protocol. Next, we present Dumbo-NG to address the challenge of latency-throughput tension by redesigning the methodology of asynchronous BFT protocols. Another benefit of the new methodology is that it can conquer the censorship threat without extra cost. Furthermore, we consider a realistic environment and present Bolt-Dumbo Transformer (BDT), a generic framework for practical optimistic asynchronous BFT to achieve the "best of both worlds" in terms of the advantages of deterministic BFT and randomized (asynchronous) BFT.