The impact of ML-KEM and ML-DSA on mTLS connection Time-to-Last-Byte
summary
Measured mTLS using ML-KEM and ML-DSA shows large certificate sizes break small TCP congestion windows and add round trips: a 22 kilobyte certificate exceeds a 15 kilobyte congestion window, adding ~52 ms extra RTT at 100 Mbps (≈50% handshake slowdown) and ~115 ms at 1 Mbps. Increasing CW to ~34 KB removes the extra RTT and reduces overhead to ~2 ms (≈2%), doubled for mTLS to ~3.2 ms; transfers of 50–150 KB see only ~1–1.4% TTLB overhead. Emulation uses tunable CW on client and server.
AI-written summary · about
chapters (16)
- 0:00Welcome
- 1:05Speaker overview
- 2:28Postquantum migration importance
- 3:31mTLS in network security
- 6:22Postquantum mTLS latency
- 8:33Classical vs postquantum RTT
- 12:02Congestion window impact
- 14:28mTLS handshake results
- 18:48Large windows performance
- 20:13Data transfer latency
- 22:21Conclusion and key takeaways
- 24:08Q&A on certificates
- 25:12Q&A on ML-KEM/ML-DSA sizes
- 26:16Q&A on zero RTT
- 26:42Q&A on congestion window
- 27:47Q&A on data references
related talks
20mmmTLS: Scaling the Performance of Encrypted Network Traffic Inspection
43mMathematics and development of fast TLS handshakes
30mMathematics and development of fast TLS handshakes
53mHansel and Gretel do TLS
52mDeath, Taxes, and HTTP
38mIoT TLS: Why It's Hard
30mAccelerate TLS with DPDK
56mThe first 200 milliseconds of HTTPS