Ambulance: Saving BFT through Racing
Talkat 20th USENIX Symposium on Operating Systems Design and Implementation (OSDI '26) · 2026-07-13..07-15·
summary
Ambulance replaces timeouts with protocol-rigged races where the leader runs a two-step path and others a three-step path so the leader has a one-step head start; leaders form sports car certificates, proposers form truck certificates and trucks are reused to speed recovery. A Rust prototype across four EC2 regions matches Autobahn’s common-case throughput and latency, cuts P99 in production by ~2x, and reduces peak slowdown latency by 2.2x–3x versus aggressive or conservative timeout settings.
AI-written summary · about
chapters (9)
- 0:00BFT consensus basics
- 2:27Slowdowns break consensus
- 3:51Timeouts in failure detection
- 5:55Hedging slowdown limits
- 7:20Protocol racing detector
- 8:42Ambulance design goals
- 9:24Ambulance race phase
- 12:56Ambulance evaluation
- 15:45Conclusion and future
related talks
35mMove fast, build things… Safely!
15mAlea-BFT: Practical Asynchronous Byzantine Fault Tolerance
34mDriving Safety Forward: Lessons Learned From Deploying OSS in Real-world Automotive
55mScaling out operational research with Rust
17mBFTBrain: Adaptive BFT Consensus with Reinforcement Learning
57mReliability Nirvana
15mBoosting System Resilience: Exponential Backoff
27mBulletproof API's