Back to Docs
Distributed Systems

Consensus Algorithms

How distributed nodes agree on a single source of truth without trusting each other. From Paxos to Raft to PBFT — the recipes that keep systems consistent under failure.

P

Paxos

The classic. Proposers, acceptors, learners. Hard to implement correctly but foundational to every consensus system built since.

R

Raft

Designed for understandability. Leader election, log replication, safety. Powers etcd, Consul, and countless production clusters.

B

PBFT

Practical Byzantine Fault Tolerance. Tolerates malicious nodes, not just crashes. The backbone of permissioned blockchain networks.

N

Nakamoto

Proof-of-work consensus without identities. Longest chain wins. Bitcoin's breakthrough — probabilistic finality at global scale.

Choosing the right recipe

Crash fault only? Raft or Paxos. Raft if your team values debuggability. Paxos if you need multi-paxos for geo-distribution.

Byzantine actors? PBFT for small (<100) known validators. Tendermint for BFT with rotating validator sets. Nakamoto for open participation.

Performance sensitive? Leader-based protocols (Raft, Multi-Paxos) handle high throughput. BFT variants add rounds — expect 2-3x latency overhead.

Ready to dive deeper?

Explore our distributed locks recipe for practical consensus patterns in production.