Researchers propose probabilistic consensus to replace 50-year-old distributed systems model

A team from HotOS '25 proposes replacing deterministic consensus protocols with probabilistic models that better reflect real-world uncertainty. Traditional systems rely on rigid failure thresholds (f-machines), but the new approach uses individual machine failure curves to bypass bottlenecks like majority quorum intersection. This enables more reliable, efficient, and cost-effective distributed systems by acknowledging that real-world failures are probabilistic, not binary. The paper argues this could revolutionize cloud infrastructure and database design.