← Lessons

quiz vs the machine

Gold1400

Concurrency

Consensus in Practice Tradeoffs

Choosing among Paxos, Raft, and BFT variants based on real engineering constraints.

6 min read · core · beat Gold to climb

No single best protocol

The consensus algorithms differ less in raw capability than in engineering tradeoffs. Choosing one means weighing fault model, latency, throughput, and how understandable the protocol is to the team operating it.

The main axes

  • Fault model: crash tolerant protocols like Paxos and Raft assume nodes only stop. Byzantine protocols like PBFT and Tendermint also tolerate lies but need 3f plus 1 nodes and more messages.
  • Understandability: Raft was explicitly designed to be teachable, which lowers operational risk versus the famously subtle Paxos.
  • Latency and load: a single leader simplifies reasoning but bottlenecks throughput. Leaderless designs like EPaxos spread load at the cost of complexity.
  • Quorum tuning: Flexible Paxos lets you trade fast commits against costlier recovery.

Practical guidance

Most internal datacenter systems pick Raft or Multi Paxos because crash faults dominate and operability matters most. Open or adversarial settings, like public blockchains, justify the heavier Byzantine protocols. The right choice follows the threat model and the team rather than any abstract notion of the best algorithm.

Key idea

Consensus protocol choice is a tradeoff among fault model, latency, throughput, and understandability, so match the protocol to your threat model and operating team.

Check yourself

Answer to earn rating on the learn ladder.

1. When are heavier Byzantine protocols justified?

2. Why do many datacenter systems pick Raft?