Building Probabilistic BFT Protocols Through Subsampled Voting
summary
Proposes a family of Avalanche DAG probabilistic BFT protocols that use repeated random subsampling (≈20 peers per query) and an increasing supermajority check (≈80%) to reach binary or total order. Claims log n convergence (under tests <17 rounds), amortized O(1) message cost, safety under asynchrony and liveness under synchrony, and a parametrized Byzantine tolerance (≈20% for chosen parameters). Shows a real 2000-node C++ AWS deployment achieving ~200 ms avg finality and reports a security target of one failure per 20,000 years at 10k TPS.
AI-written summary · about
chapters (9)
- 0:00Introduction to protocols
- 0:59Classical vs Nakamoto
- 3:31Current industry solutions
- 5:14The proposed mechanism
- 7:29Stochastic consensus
- 9:40Adversarial model
- 10:49Avalanche implementation
- 12:20Achieving total ordering
- 14:48Q&A session
related talks
15mAlea-BFT: Practical Asynchronous Byzantine Fault Tolerance
01h 07mIndustry Presentations
18mChitu: Avoiding Unnecessary Fallback in Byzantine Consensus
17mBFTBrain: Adaptive BFT Consensus with Reinforcement Learning
30mDHT, BFT, OMG: Know Your Building Blocks for Decentralized Applications
42mStochastic reasoning in decentralised systems
16mDispersedLedger: High-Throughput Byzantine Consensus on Variable Bandwidth Networks
16mShoal++: High Throughput DAG BFT Can Be Fast and Robust!