Reproduction Package for RepliGuard: Policy-Driven Replica Management Framework for Protecting against Acoustic Attacks

Three different artifacts are zipped into a single file due to file limit. Please unzip the artifacts_package.zip and refer to README.md (juypter notebook for data and graph plots) for review. CRDB run scriptsThis artifact provides the shell scripts used to conduct I/O degradation experiments on a 4-node CockroachDB cluster. It includes scripts for injecting configurable I/O latency via Linux dm-delay on a victim node, running CockroachDB kv workloads, collecting Prometheus metrics, and applying a partial replica migration mitigation strategy using CockroachDB zone configurations and range splits. Intended to reproduce the experimental results reported in the paper. data and graph plotsThis artifact contains the raw experimental data (TSV files) collected from Prometheus during attack and mitigation runs, along with a Jupyter notebook (plotting.ipynb) and helper scripts used to generate all figures in the paper. Data covers single-node and multi-node CockroachDB configurations under varying levels of I/O degradation (0–40 ms injected delay). LSTM artifactThis artifact provides Python scripts and eBPF-collected I/O latency traces for training and evaluating machine-learning classifiers (LSTM and SVM) that detect I/O degradation attacks at the block device level. Input features are sub-latency components (start-to-queue, queue-to-issue, issue-to-complete) extracted from bpfLogs. The artifact includes labeled benign and attack trace data and produces trained model files along with confusion matrix metrics