How much will ML-DSA Signatures affect Web Metrics after all?
summary
Claim: ML-DSA adds about ~15KB to the TLS handshake, measured impact using WebPageTest.org and Chrome under 4G/cable/3G profiles. Results show TLS-only time can rise dramatically (example: handshake 66% slower), yet user-facing metrics move much less — TTFB <2% slower on the 4G scenario, FCP/LCP ≈1–2% slower; trimming auth data to ~10KB cuts TLS slowdown to ~5% and user metrics to ~1–1.5%. Small “lean” third‑party connections (~5–8KB) are most affected; resumption is excluded.
AI-written summary · about
chapters (13)
- 0:00Postquantum crypto session intro
- 1:26ML-DSA impact on web metrics
- 2:49Industry claims on ML-DSA overhead
- 7:13Methodology with WebPageTest
- 10:20Initial data transfer findings
- 12:51Quantified user-facing impact
- 15:17Cloudflare vs HTTP Archive data
- 17:44Logged-in page connections analysis
- 20:15Key takeaways for ML-DSA adoption
- 22:03Web design and crypto costs
- 24:29Q&A: Metric relevance
- 25:11Q&A: Parameter sets
- 26:15Summary
related talks
53mHansel and Gretel do TLS
20mmmTLS: Scaling the Performance of Encrypted Network Traffic Inspection
53mThe Long-Tail of Performance
47mQuantum-Safe TLS in Practice With Open Quantum Safe & OpenSSL 3
41mWill Machine Learning Replace The WAF?
28mBirth of the Post-Quantum Internet
14mLoRaTLS: Approaching BSI compliances
37mMobile Web Performance