[Cryptography] Creating a Parallelizeable Cryptographic Hash Function
David Leon Gil
coruus at gmail.com
Wed Oct 8 20:09:41 EDT 2014
On Wed, Oct 8, 2014 at 12:20 PM, Bear <bear at sonic.net> wrote:
> a) you want aliasing (accidental collisions) to be minimized.
> b) you want the hash to be fast to compute.
> c) you want "random" distribution of GUID's in the hash table
> (ie, no single part of your hash table should be *more*
> full than the other parts, or at least not by any margin
> distinguishable from statistical noise on random numbers).
As an aside, what you're describing is a function with low discrepancy
on the inputs. This is (interestingly enough) different from a
function that generates values indistinguishable from uniform random
numbers in the interval. (The values "anti-bunch", in some sense,
compared to random numbers.) See, e.g.,
http://en.wikipedia.org/wiki/Low-discrepancy_sequence for the
fascinating details of low-discrepancy sequences and quasi-random
(I used to use (t,m,s)-nets and Niederreiter sequences for statistical
work. And folks here might know Harald Niederreiter's name from his
linear-code-based public key cryptosystem.)
More information about the cryptography