Byzantine Fault Tolerance (BFT)

Byzantine Fault Tolerance describes a mathematical property, not a specific piece of software: a network has it if honest participants can still agree on the correct state of the system even when some nodes crash, go offline, or actively send conflicting or false information to different peers. The name comes from the Byzantine Generals Problem, a thought experiment about coordinating an attack when some generals might be traitors feeding contradictory orders to different parts of the army.

In practice, most BFT designs share a hard limit: consensus stays safe as long as fewer than one-third of participating nodes are faulty or dishonest, a bound proven by Castro and Liskov's Practical Byzantine Fault Tolerance (pBFT) algorithm in 1999. pBFT works by having a rotating leader propose a block and every replica exchange multiple rounds of signed messages until a supermajority agrees, giving instant finality but requiring heavy node-to-node communication that limits scale.

Blockchains adopted BFT thinking to solve the same trust problem among mutually suspicious strangers rather than known servers in a data center. Tendermint, the consensus engine behind the Cosmos ecosystem, streamlines pBFT into propose, prevote, and precommit rounds among validators, finalizing blocks once two-thirds of voting power signs off. Hyperledger Fabric uses a pBFT-style ordering service for permissioned enterprise chains, and Ethereum's Casper FFG finality gadget borrows BFT-derived safety guarantees to checkpoint blocks on the beacon chain.

The trade-off is scalability versus security: BFT systems tolerate malicious actors well but need more message overhead as validator sets grow, which is why many chains cap active validator counts or combine BFT rounds with other mechanisms to balance speed, decentralization, and safety.