Byzantine failure
A Byzantine failure permits a component to deviate arbitrarily from its specified behavior, including sending conflicting information to different participants.
The model describes behavior, not intent: defects and malicious actions can both cause arbitrary deviations. Equivocation means conflicting statements about the same protocol decision; it is not a synonym for all Byzantine failures. [1] [3]
Authentication identifies an origin, not truth. A faulty replica can sign conflicting statements using its own key. PBFT's basic client accepts matching authenticated replies from f + 1 distinct replicas for the same request, with at most f faulty replicas. [1] [3]
PBFT's 3f + 1 configuration is model-specific. Safety and liveness have different assumptions; liveness depends on communication timing. Consistent execution cannot make an authorized client's bad write useful, and a faulty replica may disclose information. A missing reply alone does not establish a Byzantine fault. [2]
Signed-message agreement can use a different fault bound. SM(m) assumes reliable delivery, detectable message absence and unforgeable signatures. Signature forwarding exposes conflicting orders; signatures alone do not supply the timing assumptions. This synchronous agreement result is not a replacement replica count for PBFT. [4]
Relations
- related
- Fault model, Consensus, Quorum intersection
Limitations
- A Byzantine-fault-tolerant protocol's guarantees depend on its stated faulty-replica bound, authentication and progress assumptions; the term itself does not certify a deployment. [1, 2]
Citations
- Castro and Liskov. Practical Byzantine Fault Tolerance. OSDI 1999, section 2: System Model.
- Castro and Liskov. Practical Byzantine Fault Tolerance. OSDI 1999, section 3: Service Properties.
- Castro and Liskov. Practical Byzantine Fault Tolerance. OSDI 1999, sections 4.1-4.4: client replies, normal operation and view changes.
- Lamport, Shostak and Pease. The Byzantine Generals Problem. TOPLAS 4(3), 382-401, 1982; sections 3, 4 and 6.
Further reading
Sources & caveats
- Primary-source synthesis of a failure-model concept; not a new consensus protocol, cluster experiment or estimate of failure frequency.