BENCHMARKS.md

Benchmarks

This project ships reproducible performance and regression gates for Sentinel proxy overhead.

Quick Run

npm run benchmark -- --duration 3 --connections 16 --pipelining 1
npm run benchmark:gate
npm run benchmark:datasets

Reliability stress/chaos path:

npm run reliability -- --websocket-requests 8

Published Benchmark Artifacts

• Methodology: docs/benchmarks/METHODOLOGY.md
• Competitor comparison: docs/benchmarks/COMPETITOR_COMPARISON.md
• Sentinel benchmark snapshot: docs/benchmarks/results/sentinel-v4.json
• OWASP coverage matrix data: docs/benchmarks/results/competitor-coverage.json
• Standard adversarial datasets snapshot: docs/benchmarks/results/standard-datasets.json
• Standard dataset fixtures: docs/benchmarks/datasets/*.json

Current Sentinel Baseline (published)

• direct p95: 33 ms
• sentinel p95: 34 ms
• p95 overhead: 1 ms (3.03%)
• throughput delta: -57 req/sec

Source: docs/benchmarks/results/sentinel-v4.json

Standard Adversarial Dataset Baseline

• datasets: advbench-mini, trojai-mini
• total cases: 30
• expected-detection cases: 24
• overall detection rate: 100% (threshold 85%)
• false positives: 0

Source: docs/benchmarks/results/standard-datasets.json

Regression Policy

• Benchmark gate fails CI on configured regression thresholds (npm run benchmark:gate).
• Separate perf gates cover P0/P1/P2/P3/V4 engine groups.
• Reliability suite validates timeout, circuit-breaker, chaos, and websocket paths.

Notes

• Run with NODE_ENV=production for realistic behavior.
• Use comparisons from the same runner profile when evaluating regressions.
• Competitor metrics are only asserted when reproducible in-repo; otherwise marked not_measured.