News for hackers. Stuff that matters.
Re:Hash Functions all the way down by aappleby on Wednesday August 03, 2016 @05:40PM (#12221768)
It'd be interesting to see how the 128-bit Murmur versions compare.

Something not mentioned in the "Quality" section is that FNV and djb2 produce may produce decent results on words/filenames/source code, but they can fail catastrophically (produce far more collisions than expected) on certain patterns of keys - SMHasher is designed not just to check performance, but to try and find patterns of keys that 'break' a hash function.

Murmur3 tries to be short, simple, portable, and efficient on small keys while still passing SMHasher. xxHash, Spooky, and the Farm/City/Metro suite all use a lot of loop unrolling and/or platform-specific optimizations to get the highest possible performance when hashing large chunks of data.

-Austin (author of Murmur and SMHasher)

Re:Hash Functions all the way down by ComNik on Wednesday August 03, 2016 @06:42PM (#12222076)
Thank you for Murmur3! I have been using it in a consistent hashing implementation as part of my bachelor's thesis and it is performing very well.
Re:Hash Functions all the way down by aras_p on Wednesday August 10, 2016 @03:33AM (#12260068)
> It'd be interesting to see how the 128-bit Murmur versions compare.

Added them (and some more hash functions) in a follow-up post: http://aras-p.info/blog/2016/08/09/More-Hash-Function-Tests/

Re:Hash Functions all the way down by _dps on Thursday August 04, 2016 @04:25AM (#12224213)
First, thanks for all your work on Murmur and SMHasher!

I wonder if you have any recommendations for a one-byte-at-a-time hash that survives SMHasher? I've been using Jenkins OAAT and haven't bumped into any problems but I've seen conflicting reports about its robustness. Any thoughts on Jenkins OAAT or an alternative one-at-a-time hash would be appreciated :)

Re:Hash Functions all the way down by aappleby on Thursday August 04, 2016 @04:12PM (#12228118)
If you can only hash data a byte at a time, you have bigger problems. ;)

I have no real suggestions, but it would be trivial to feed Murmur one byte per loop iteration and then ditch the 'tail' section.

Re:Hash Functions all the way down by aras_p on Thursday August 04, 2016 @01:32AM (#12223596)
Yeah, should revisit this and test more hash functions. Just ran out of time I wanted to spend on this (hey, I got some actual work to do too :))