LMS: Lighter, faster key generation
summary
Describes an implementation that combines SIMD SHA‑256 hashing and multithreading with an adapted RFC8554 low‑memory iterative algorithm to cut LMS key‑generation work by roughly the number of SIMD lanes (example: 4× parallel lane hashing) and avoid storing full levels. Concrete numbers: a common LM‑OTS leaf needs ~9,000 sequential hashes, the largest LMS tree (height 25) previously cost gigabytes of storage but the small‑memory scheme reduces that to about 2 GB → ~1 MB. The approach generates keys and roots faster and keeps signing overhead to at most H/2 additional hashes.
AI-written summary · about
chapters (12)
- 0:00LMS key generation intro
- 1:52LMS algorithm and optimizations
- 2:57Hashing requirements
- 7:15Leaf calculation process
- 9:30Parallelizing hashing with SIMD
- 11:39Efficient root computation
- 13:47SIMD root computation
- 16:37LMS tree node lifetime
- 17:20Small memory LMS schemes
- 19:33State management in LMS
- 20:15Key generation optimizations
- 21:19Conclusion and Q&A
related talks
24mUpdate on end-to-end PKI and HSM integrations with ML-DSA
46mAchieving Quantum Readiness: A Comprehensive Approach
15mLXM: Better Splittable Pseudorandom Number Generators (and Almost as Fast)
01h 40mHistorical Talk on Lattice-Based Cryptography
01h 04mThe Mathematics of Lattices II
44mIP Protection and Privacy in LLM: Leveraging Fully Homomorphic Encryption
56mSymmetric-Key Primitives
40mImplementation