Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

smhasher results without the Rust version yet (which should be ~2x faster):

http://rurban.github.io/smhasher/doc/table.html

It's of course much faster as most of the other crypto hashes, but not faster than the hardware variants of SHA1-NI and SHA256-NI. About 4x faster than blake2.

Faster than SipHash, not faster than SipHash13.

The tests fail on MomentChi2 dramatically, which describe how good the user-provided random seed is mixed in. I tried by mixing a seed for IV[0], as with all other hardened crypto hashes, and for all 8 IV's, which didn't help. So I'm not convinced that a seeded IV is properly mixed in. Which is outside the usage pattern of a crypto or digest hash (b3sum), but inside a normal usage.

Rust staticlib is still in work, which would parallize the hashing in chunks for big keys. For small keys it should be even a bit slower. b3sum is so much faster, because it uses many more tricks, such as mmap.



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: