Skip to content
Science of Blockchain Conference 2026

Hydrangea: Optimistic Two-Round Partial Synchrony with Improved Fault Resilience

Session summary

The presenter introduces Hydrangea, a Byzantine fault tolerant state machine replication protocol that targets optimistic two‑round commit latency under a mixed fault model with both Byzantine and crash failures. Working in a partially synchronous network, Hydrangea aims to preserve safety at all times and liveness after network stabilization, while navigating the known trade‑off between latency and fault resilience. The authors argue that realistic systems experience many crash‑type failures and therefore model the system with parameters for Byzantine faults (F), crash faults (C), and an additional redundancy parameter (K). Under this framework, they show that certain prior mixed‑model designs can violate safety under partial synchrony when leaders equivocate.

Hydrangea’s key idea is to enforce stronger quorum intersections between voting and view‑change quorums, ensuring that blocks with more honest support are extended even when an adversary delays messages or misreports past votes. In its base configuration with N = 3F + 2C + 1, Hydrangea offers an optimistic fast path tolerating up to P = C/2 faults in two rounds and a slow path that commits in three rounds while tolerating F Byzantine and C crash faults. The authors extend this to a more general N = 3F + 2C + K + 1 setting, derive near‑tight lower bounds on achievable optimistic resilience, and report experimental results indicating lower latency than a three‑round baseline under geo‑distributed conditions.

Key takeaways

  • Hydrangea operates under a mixed Byzantine‑and‑crash fault model in a partially synchronous network and seeks optimistic two‑round commits without sacrificing safety.
  • The protocol strengthens quorum intersection between voting and view‑change phases, and in its basic setting with N = 3F + 2C + 1 it supports an optimistic fast path up to P = C/2 faults and a three‑round slow path for up to F Byzantine and C crash faults.
  • The authors generalize to N = 3F + 2C + K + 1, prove a near‑tight lower bound on the maximum optimistic fault threshold P, and experimentally observe latency improvements over a three‑round Byzantine baseline in geo‑distributed tests.