Igor,

If you look at the docs, it is supposed to have excellent collision
resistance.
See http://tanjent.livejournal.com/756623.html

Given that cache objects are supposed to have a limited shelf life
(days, months), I would think that collision resistance is vs. hashing
performance is a decent tradeoff.

I guess if you put in an option to specify the hash, that would work
too. FNV and Murmur have been adopted on various open source projects as
optional hashs because of the hashing performance.


--J

Igor Sysoev wrote:
> On Sat, Apr 18, 2009 at 08:27:11PM +0200, Joe Bofh wrote:
>
>>
>> See
>> http://search.cpan.org/~tmaesaka/Digest-MurmurHash-0.10/lib/Digest/MurmurHash.pm#BENCHMARK
>> http://murmurhash.googlepages.com/
>>
>> for sample implementations.
>
> Thank you for information, but as I understand this hash produce 32- or
> 64-bit hash. I believe it should have more collisions as compared to
> md5,
> which is 128-bit hash.

--
Posted via http://www.ruby-forum.com/.