Paxos vs Raft: Have we reached consensus on distributed consensus?
summary
Paper compares Paxos and Raft by creating a raftified Paxos variant and finds the algorithms are identical in the steady-state performance and diverge only in the leader election phase. Paxos requires log promotion and can ship large log blobs during elections; Raft uses last-log checks and avoids bulk transfers, so Raft tends to recover leaders faster out of the box. Use Raft for simple, reliable defaults; use Paxos if you will apply decades of specialized optimizations.
AI-written summary · about
chapters (18)
- 0:00Start
- 5:22nachine replication
- 7:43Paxos vs Team Raft
- 8:17blue corner, Paxos
- 20:17evel approach to consensus
- 24:32we handle follower failures?
- 30:28we handle leader failures?
- 31:54leader election algorithm
- 33:58Simple example
- 35:34A less straightforward example
- 40:36we know that Paxos & Raft are safe?
- 41:14ng safety within terms
- 42:06ng safety across terms
- 43:51Paxos & Raft differ?
- 49:39we define best?
- 50:34algorithm is more understandable?
- 52:06algorithm is more efficient?
- 56:35algorithm is the best?
related talks
40mConsensus Algorithms in Distributed Systems
51mDo Not Reimplement Distributed Consensus Algorithms
01h 32mHow to Reach Consensus with Paper, Pencil and the Raft Algorithm
47mUnderstanding Raft Consensus
01h 27mHow to Reach Consensus Using Paper, Pencil and the Raft Algorithm
47mDistributed Systems: When Democracy and Computer Science Come Together
41mThe Raft Consensus Algorithm
43mAll Together Now: An Introduction to Distributed Consensus