A fault-tolerant banking application using Linear PBFT for state machine replication in the presence of Byzantine faults. Linear-PBFT replaces PBFT’s quadratic all-to-all phases with a collector-based linear communication pattern, while keeping PBFT view-change to safely recover from faulty leaders.

• GitHub: Linear PBFT
• Demo: TODO (optional)

Notes

What I built

Replicated banking state machine

• Transfer (write) operations that go through consensus
• Balance (read-only) operations that can be served via a replica fast-path with quorum replies

Linear-PBFT normal case

• Leader assigns order (sequence numbers) and coordinates progress
• Replicas send signed votes to a collector; the collector aggregates a quorum (n−f signed messages) and broadcasts them in one message to all replicas to keep communication linear

PBFT view-change

• On suspected leader failure (timeouts), replicas exchange view-change messages
• The next leader forms a new-view after collecting 2f+1 view-change messages, ensuring safety across views

Threat model / faults tolerated

• Byzantine replicas and clients, assuming cryptography holds (messages are signed and validated)

Tested adversarial behaviors, including:

• Invalid signature / authentication failures
• Crash / “no-send” behaviors (leader or backups)
• Selective omission (“in-dark”) toward honest replicas
• Timing delays by a malicious leader
• Equivocation (conflicting ordering info to different replica subsets)

Debuggability / observability

Terminal tooling to inspect:

• per-node request logs (PrintLog)
• datastore state (PrintDB)
• per-sequence status across nodes (PrintStatus: PP/P/C/E/X)
• view history (PrintView: all new-view messages observed)