BEWARE

This deposit contains the research artifact accompanying the accepted paper "Robust and Automated Reconfiguration of Byzantine Wide-Area Replication" (DSN 2026, to appear). The artifact includes source code, benchmark scripts, latency datasets, configuration files, and reference results used to evaluate BEWARE, a robust reconfiguration framework for Byzantine fault-tolerant state-machine replication (BFT-SMR) in wide-area networks.

BEWARE targets self-reconfiguring BFT-SMR systems in which replicas periodically adapt their configuration based on observed network latency. Existing reconfiguration approaches can improve consensus latency by assigning leader roles and voting weights according to network conditions, but they are vulnerable to Byzantine replicas that falsify latency reports, obtain influential voting weights, or delay protocol messages after reconfiguration. BEWARE addresses these limitations by combining robust latency-matrix sanitization, Byzantine-safe weighted voting, and machine-learning-assisted configuration selection.

The artifact contains two complementary codebases. First, the Python simulation framework reproduces the paper’s simulation experiments and figures. It includes implementations of virtual coordinate systems, latency sanitization methods, weighted-voting optimization, PBFT-related evaluation logic, benchmark scripts, and figure-generation scripts. Second, the Java implementation integrates BEWARE into a fork of BFT-SMaRt and supports deployment-style benchmark experiments.

Methodology and techniques used include:

Robust virtual coordinate systems for sanitizing latency matrices under Byzantine latency-reporting attacks.Clustering-based filtering and clipping of latency-derived forces to reduce the influence of falsified reports.Weighted Byzantine quorum reconfiguration for assigning replica voting weights and leader roles.Simulated annealing and differential-evolution-based configuration search.Machine-learning-assisted latency prediction to avoid configurations that perform poorly under adversarial behavior.Simulation and benchmark evaluation using wide-area latency datasets and emulated Byzantine behaviors such as latency poisoning and message delays.Java/BFT-SMaRt deployment-style evaluation using configurable replica settings and benchmark scripts.

The repository contains Python source code, Java source code, shell scripts, Gradle configuration, benchmark CSV files, serialized latency datasets, Java archive dependencies, configuration files, and Markdown documentation. Python dependencies are listed in requirements.txt. The Java implementation can be built using the included Gradle wrapper.

Data and provenance:

The included datasets are network-latency datasets used for simulation and emulation of wide-area deployments, including WonderNetwork-derived latency data. The artifact also includes reference benchmark outputs used to reproduce or verify the paper’s figures. No human-subject data is included. The experiments use network measurements, simulations, and emulated Byzantine behavior; no interaction with private users or private systems was performed for the included data collection.

Ethics and legal compliance:

This artifact does not contain personal data or human-subject research data. The included cryptographic key material in the Java/BFT-SMaRt configuration is intended only for local experimental and reproducibility purposes and must not be used in production systems.

Licensing:

The top-level artifact code is released under the MIT License unless otherwise stated. The Java implementation under src/java/bftsmart-beware/ is derived from BFT-SMaRt and retains its own Apache-2.0 license and documentation. Users should consult nested README and LICENSE files for details on third-party or derived components.