Failure Mode Spectrum¶

Distributed system failures exist on a spectrum from benign (crash) through omission (lost messages) to malicious (Byzantine), with handling complexity increasing dramatically - most systems optimize for crash-stop failures because Byzantine tolerance is expensive.

Tension Addressed¶

Simplicity vs. Fault Tolerance

Recognizable As¶

Crash-consistent vs. Byzantine-fault-tolerant systems; 2f+1 vs. 3f+1 replica requirements; heartbeat/health check mechanisms; trusted vs. untrusted environments

Canonical Sources¶

Lamport, Shostak & Pease, "The Byzantine Generals Problem" (1982); Schneider, "Implementing Fault-Tolerant Services Using the State Machine Approach" (1990)

Transfers To¶

Blockchain consensus, financial systems, critical infrastructure, any system with adversarial participants

Sub-concepts: - Crash-stop failure: Process halts and never recovers; simplest to handle - Omission failure: Messages lost; requires acknowledgment and retry - Byzantine failure: Arbitrary behavior including malicious; requires 3f+1 replicas to tolerate f failures