DepChain — Permissioned Blockchain with BFT Consensus

DepChain is an end-to-end permissioned blockchain system I built for the Highly Dependable Systems course at IST. The system tolerates f = ⌊(n-1)/3⌋ Byzantine replicas while still settling transactions and executing arbitrary smart contracts.

What I built

Consensus layer (Basic HotStuff over UDP)

I implemented the Basic HotStuff Byzantine Fault Tolerance algorithm from scratch on top of unreliable networks:

• A layered link stack: FairLossLink → StubbornLink → AuthenticatedPerfectLink over UDP.
• Authenticated Perfect Links with an ephemeral X25519 Diffie-Hellman handshake (signed with Ed25519) deriving per-session HMAC keys.
• Quorum Certificates secured with BLS threshold signatures via the JPBC pairing-based crypto library, so a single 48-byte signature attests 2f+1 votes.
• Three-phase HotStuff (PREPARE → PRE-COMMIT → COMMIT → DECIDE) with leader rotation and view changes.

Transaction processing engine

On top of consensus I built a full transaction engine:

• Genesis loading with deterministic world-state snapshot/restore, so a crashed replica recovers to a consistent state.
• Mempool ordered by gas price, nonce, and gas cap.
• EVM execution integrated through Hyperledger Besu, executing arbitrary Solidity bytecode with a custom gas-fee mechanism.
• DepCoin native transfers with overspend, double-spend, and replay protection.
• Block persistence with crash-recovery round-trip tests.

IST Coin (frontrunning-resistant ERC-20)

I designed and shipped an ERC-20 token whose approve() function is redesigned to prevent the classic Approval Frontrunning attack — a known problem with naive ERC-20 implementations where an attacker can front-run a balance change to an allowance.

Test surface

The project ships with 45 JUnit 5 tests covering:

• Link-stack correctness (stubborn delivery, perfect-link rejection of wrong keys)
• HotStuff data structures (TreeNode, Message, QC)
• Crypto primitives (Ed25519, BLS threshold, QC integration)
• Full 4-replica consensus over UDP
• Crash tolerance and silent-Byzantine fault injection
• Spoofing, overspend, replay, double-spend, frontrunning attack scenarios

Why it matters

Building DepChain meant doing distributed-systems engineering across the whole stack: cryptography, networking, consensus, virtual-machine integration, and adversarial testing. The result is a system that’s not just “it compiles” but actually demonstrably tolerant to crashes and Byzantine peers under fault injection.