ビザンチン耐障害性（BFT）

Byzantine Fault Tolerance addresses the “Byzantine Generals Problem,” a thought experiment where loyal generals of a Byzantine army must agree on a common battle plan while communicating through potentially treacherous messengers. Some generals may be traitors and actively try to foil the plan by sending conflicting information. BFT algorithms provide a solution to this problem in distributed computing, ensuring that all non-faulty (loyal) nodes reach the same conclusion, despite the presence of faulty (traitorous) nodes that can lie, collude, or behave arbitrarily.

The seminal 1982 paper by Lamport, Shostak, and Pease proved that for a system to tolerate [latex]f[/latex] Byzantine faults, there must be at least [latex]3f+1[/latex] total nodes (or generals). This means that to tolerate one malicious actor, a system needs at least four nodes in total. The core of BFT algorithms involves multiple rounds of message passing among the nodes. In each round, nodes exchange their current state or proposed value, along with the messages they have received from other nodes. This allows loyal nodes to cross-check information and identify inconsistencies, eventually converging on a single, correct value that is agreed upon by the majority.

Early BFT algorithms, like Practical Byzantine Fault Tolerance (PBFT), were computationally expensive and had high communication overhead, limiting their scalability. However, their ability to provide safety and liveness in an adversarial environment was revolutionary. The rise of blockchain technology in the late 2000s created a massive new application area for BFT. Consensus mechanisms in cryptocurrencies like Bitcoin (using Proof-of-Work, a probabilistic form of BFT) and newer Proof-of-Stake systems (often using explicit BFT protocols like Tendermint or HotStuff) rely on these principles to ensure the integrity and immutability of the distributed ledger, even if a significant fraction of network participants are malicious.