An Analysis of Avalanche Consensus

TL;DR

The paper formalizes the performance and security of Snow consensus protocols underlying Avalanche, proving a fundamental latency bound of ${\Omega}(\frac{\log n}{\log k})$ rounds and an upper bound of ${O}(\log n)$ under a ${O}(\sqrt{n})$-bounded adversary for Slush and its Snowflake/Snowball successors. It identifies a tension between achieving negligible failure probability (via the security parameter ${\beta}$) and fast finalization, showing an inherent trade-off that cannot be overcome within the original Snowball/Snowflake design. To reconcile security and latency, the authors introduce Blizzard, a Slush-like protocol with a confidence-based termination that guarantees consensus in ${O}(\beta + \log n)$ rounds with all-but-negligible probability. The work provides concrete guidance on parameter choices (e.g., minimal ${\alpha}$ for a given ${k}$) and demonstrates practical improvements for Avalanche-style blockchains by reducing latency while maintaining rigorous probabilistic guarantees.

Abstract

A family of leaderless, decentralized consensus protocols, called Snow consensus was introduced in a recent whitepaper by Yin et al. These protocols address limitations of existing consensus methods, such as those using proof-of-work or quorums, by utilizing randomization and maintaining some level of resilience against Byzantine participants. Crucially, Snow consensus underpins the Avalanche blockchain, which provides a popular cryptocurrency and a platform for running smart contracts. Snow consensus algorithms are built on a natural, randomized routine, whereby participants continuously sample subsets of others and adopt an observed majority value until consensus is achieved. Additionally, Snow consensus defines conditions based on participants' local views and security parameters. These conditions indicate when a party can confidently finalize its local value, knowing it will be adopted by honest participants. Although Snow consensus algorithms can be formulated concisely, there is a complex interaction between randomization, adversarial influence, and security parameters, which requires a formal analysis of their security and liveness. Snow protocols form the foundation for Avalanche-type blockchains, and this work aims to increase our understanding of such protocols by providing insights into their liveness and safety characteristics. First, we analyze these Snow protocols in terms of latency and security. Second, we expose a design issue where the trade-off between these two is unfavorable. Third, we propose a modification of the original protocol where this trade-off is much more favorable.

Figures (1)

• Figure 1: Plots of $\delta(p_i)$ for different parameters $k$ and $\alpha$. For $k=2\alpha\!-\!1$ the expected progress for larger $\alpha$ dominates those for smaller (note that in the extreme case $k=\alpha=1$ there is no expected progress). For fixed $k$ the opposite is true. The combination $k=20, \alpha =15$ was suggested by the whitepaper Rocket2019. Note that $\delta(p_i)$ is point-symmetric with respect to the point $(\tfrac{1}{2},0)$.

Theorems & Definitions (67)

• Definition 2.1: State of Stable Consensus
• Definition 2.2: Consensus Problem
• Definition 2.3: $F$-Bounded Adversary
• Definition 2.4: Negligible Probability, Security Parameter
• Lemma 3.1
• proof
• Lemma 3.2
• proof
• Lemma 3.3
• proof