Phase Coupling in Cross-Chain Atomic Swaps: A Predictive Latency Model and the Cost of Trustlessness on Hyperledger Besu and Fabric
Description
This dataset contains the complete testbed configuration, code and raw measurements for an empirical comparison of two ways of moving an asset between a Hyperledger Besu network (four validators, QBFT consensus, EVM) and a Hyperledger Fabric channel (two organisations, Raft ordering): a hash time-locked contract (HTLC) atomic swap, which needs no trusted party, and a custodial relayer, which holds keys on both chains. Contents: - results/ — 26 raw JSON result files, one per experiment run: single-chain latency and load sweeps for both chains; decomposed atomic swaps at block periods of 5 s, 4 s and 3 s; a paired, interleaved comparison of 50 atomic swaps against 50 custodial transfers; failure-semantics tests (counterparty abandonment, relayer outage); and validator crash-fault runs at round timeouts of 10 s and 15 s. Runs made from November 2026 also record the number, timestamp and proposer of every Besu block. - vm1-besu/ — the Besu genesis with all block-period transitions, Docker Compose file, Solidity contracts (HashedTimelock.sol, Bridge.sol) and the deployment and benchmark scripts, exactly as run. - vm2-fabric/ — the HTLC chaincode (chaincode as a service), the Fabric benchmark scripts and the cross-chain orchestration, experiment and analysis scripts, exactly as run. - figures/ — the five figures of the associated article, regenerated from results/. - Documentation: README, SETUP, RUNBOOK, RESULTS, TROUBLESHOOTING, config-history (every configuration change during the campaign, with block numbers and times) and a results schema. Principal findings that can be recomputed from the files: removing the trusted party cost 2971 ms per transfer (95% CI ±188 ms; Welch's t(49.3) = 31.0), a factor of 2.40; the cross-chain preimage read took 9.5 ms; a phase-coupling model predicts the latency of the dependent Besu stage within 1.6% at three block periods; under a single validator crash, each swap paid one full round timeout or none. Private keys are not included. The code is laboratory software and must not be used on a network that holds value. Associated article: I. Ahmad and O. Farooq, "Phase Coupling in Cross-Chain Atomic Swaps: A Predictive Latency Model and the Cost of Trustlessness on Hyperledger Besu and Fabric", submitted to the Journal of Parallel and Distributed Computing.
Files
Steps to reproduce
1. Hardware: three virtual machines on one host (the original used VMware ESXi 8 on 2x Xeon E5-2620 v4, 64 GB RAM). VM1 (at least 4 vCPU, 8 GB) for Besu; VM2 (at least 4 vCPU, 8 GB) for Fabric and the orchestrator; VM3 (optional) for Prometheus and Grafana. Ubuntu 24.04 or 26.04, Docker 29, Node.js 22. 2. VM1: start the four-validator Besu QBFT network from vm1-besu/besu-qbft (genesis.json, docker-compose.yml); deploy HashedTimelock.sol and Bridge.sol with the scripts in vm1-besu/blockchain-training/scripts (Hardhat, Solidity 0.8.28, evmVersion shanghai). Block period 5 s, requesttimeoutseconds 10. 3. VM2: start the Fabric 2.5.10 test network (two organisations, one Raft orderer, BatchTimeout 2 s, MaxMessageCount 10); deploy the asset-transfer-basic sample and the HTLC chaincode in vm2-fabric/htlc-cc as chaincode-as-a-service; install the scripts in vm2-fabric/fabric-bench and vm2-fabric/swap (npm install). 4. Run the experiments in RUNBOOK.md (E1 to E9): single-chain latency and load (bench-besu, load-besu, bench-fabric, load-fabric), decomposed swaps (batch-swap, batch-swap-p for other block periods), the paired comparison (compare.mjs), failure tests (fail-timeout, fail-relayer) and the validator fault (fail-validator-n). Each run writes a JSON file to results/. Block-period changes need a QBFT transition in genesis.json and a restart of all validators; the new period starts at the restart, not at the configured block (see config-history.md). 5. Recompute: node stats.mjs results/compare-*.json (Welch's test, Table 10); node periods.mjs <5 s file> <3 s or 4 s file> (Table 13); python3 figures.py --results results --out figures (all five figures). Absolute latencies depend on hardware; the paired differences, ratios and phase-coupling predictions should reproduce.
Institutions
- Sir Syed CASE Institute of TechnologyIslamabad, Islamabad