On Sun, Sep 27, 2026 at 07:30:02AM +0900, Kazu Yamamoto (山本和彦) via Haskell-Cafe wrote:
Hello cafe,
I have posted a new blog article on "crypton": https://kazu-yamamoto.hatenablog.jp/entry/2026/09/25/110313
Just FYI: In crypton v2.1.2, thanks to AWS's s2n-bignum library, P-256 performance now rivals that of OpenSSL. The next target is RSA.
Many thanks for this work. Much appreciated, from both a security (attention to constant-time) and performance perspective. I've rebuilt the DNSSEC/DANE survey engine to use crypton 2.1 and it is working well (though TLS is not performance critical, since the engine's performance is constrained primarily by network-latency rather than CPU, thus uses 1400 forkIO threads to amortise an average latency in excess of 100ms). Just one nit on the README.md, the X86_64 numbers are said to be measured against OpenSSL 3.0, which is rather dated and has just reached end-of-life. The fair comparison is head-to-head, so against either 4.0 (released in April) or the "master" branch. Yes, that could mean building from source. The now outdated 3.0 release had notable performance regressions, being a major redesign of the OpenSSL core for modularity, and much subsequent effort went into streamlining its performance in later releases. Yes, the assembly modules and base cryptography algorithms (once fetched and used to perform multiple operations) may have been less affected in various use cases, but it is now prudent to use 4.0 or later when making measurements. -- Viktor. 🇺🇦 Слава Україні!