Hydra: Support Dynamic BFT with Weaker Assumptions and Explicit Request Handling

This paper presents Hydra, a dynamic BFT protocol that allows replicas to join and leave the system dynamically. It addresses the limitations of traditional static BFTs in managing membership changes and can be used to simplify the implementation of many features in modern blockchain applications. Hydra relies on weaker assumptions to achieve standard properties compared to the existing solution Dyno and introduces a configuration auto-transition protocol to ensure liveness. Through temporary configurations and explicitly defined replica responsibilities for request handling, Hydra pipelines membership requests alongside regular requests and realizes clarity, achieving a more efficient and smoother configuration transitions. It also employs a non-blocking configuration discovery mechanism, enabling new replicas to participate in consensus quickly. We formally prove Hydra's correctness under the dynamic BFT model. Experimental results demonstrate Hydra's ability to maintain throughput fluctuations within 5% during various replica join and leave scenarios, outperforming Dyno and existing BFT system supporting reconfiguration in both stability and efficiency. Hydra effectively manages scenarios that Dyno circumvents with stronger assumptions and quickly restores throughput to normal levels.