Ambulance BFT protocol races replicas to fix slowdowns
A new BFT protocol ditches timeouts for protocol-rigged races to maintain latency and throughput.
A team of seven researchers from leading institutions (including Cornell, UT Austin, and University of Stavanger) have published a new paper titled "Ambulance: saving BFT through racing" on arXiv. The paper addresses a fundamental trade-off in practical Byzantine Fault Tolerant (BFT) state machine replication deployments: timeouts. Aggressive timeouts trigger costly leader changes, while conservative timeouts leave the system idle, inflating latency. Existing alternatives—hedging (adds time-based delay) and cooperative asynchronous protocols (high common-case latency, low throughput)—each suffer from their own drawbacks.
Ambulance sidesteps this trade-off entirely by introducing protocol-rigged races. Instead of racing against a clock, replicas race against each other by executing protocol steps. This design enables Ambulance to match the high throughput and low latency of state-of-the-art timeout-based BFT while achieving the robustness of cooperative asynchronous approaches. The protocol eliminates the need for timeouts altogether, making it resilient to slowdowns without sacrificing performance. For distributed systems engineers, this means a significant leap forward in building fast, reliable fault-tolerant systems—especially for latency-sensitive applications like blockchain consensus, cloud databases, and critical infrastructure.
- Ambulance replaces timeouts with protocol-rigged races between replicas, avoiding expensive leader changes and idle periods.
- Achieves throughput and latency comparable to timeout-based BFT while matching robustness of cooperative async protocols.
- Developed by Neil Giridharan, Shubham Mishra, Lorenzo Alvisi, Natacha Crooks, Benjamin Marsh, Hein Meling, Kartik Nayak, and Grzegorz Prusak.
Why It Matters
Eliminates the latency-throughput trade-off in BFT, enabling faster and more robust distributed systems.