Liberating Distributed Consensus
summary
Argues that Paxos is overly conservative and instead uses write-once registers and per-register quorums to guarantee safety while reducing work. Shows a three-server example where majority quorums give classic two-phase consensus but a co-located client can take a fast path and reach agreement in one round-trip, falling back to the two-phase slow path when a server fails. Claims this tolerates a minority of failures and yields much lower latency in the common case.
AI-written summary · about
chapters (9)
- 0:00Introduction and speaker intro
- 1:07The dream and limits of distributed systems
- 3:36What is distributed consensus?
- 5:44Paxos: the consensus standard
- 7:28Rethinking consensus with immutability
- 12:47How Paxos ensures safety
- 17:19Paxos quorum requirements are too strict
- 18:45One-round consensus with All aboard
- 21:38Beyond Paxos: many consensus options
related talks
48mDistributed Consensus Algorithms
40mImplementing Distributed Consensus
43mAll Together Now: An Introduction to Distributed Consensus
40mConsensus Algorithms in Distributed Systems
51mDo Not Reimplement Distributed Consensus Algorithms
23mLightning Talks on Distributed Systems
18mAdoB: Bridging Benign and Byzantine Consensus with Atomic Distributed Objects
17mNear Optimal Latency Versus Cost Tradeoffs in Geo Distributed Storage