AFAIK the improvements by better keys are relatively small. Simply
using more bits with a good random number generator can decrease the
probability of collisions quite effectively. How many additional bit's
does your "secret" scheme save (assuming equal probability of
I use 64 bits.
Anything more significantly impairs speed.
collisions), compared to high quality random keys and other standard
approaches such as those maximising the hamming distances?
I have found an improvement in maximising the Hamming distances.
I found that maximum Hamming distances in fact perform abonimably.
The "secret" is that there is such a thing as OPTIMUM Hamming distance and
that the optimum Hamming distance is NOT the maximum Hamming distance.
Another improvement is to take the problem domain into account.
You can adapt the hask keys used at any point to the position or pattern
you
are hashing and virtually guarantee very good collision protection.
But optimum Hamming distance makes a big difference to maximum Hamming
distance. That's the secret :)
Ok, but that is not so special. Nobody is regenerating keys until
eternity, and of course it is possible to use (topological) knowledge of
the domain. Anyway, you did not answer the question.