On Tue, Dec 03, 2002 at 10:45:44AM -0500, Andrew Brown wrote:
> >for actual birthdays, you need 23 people to get a greater than 50%
> >chance of a collision, not sqrt(365), which is 19 (though i don't
> >think i'd bet until you reached 28 people).  19:23 == 0.826.

Actually birthdays (or more precisely the events that lead up to
them) are actually sufficiently unrandom that the odds with 23
people are well over 50%

For a cheap hash I'd look at one of those pseudo random
number generators that do a CRC like activity - if they
used XOR they would CRC each bit of the word separately,
but they use + and 'cross pollute' the CRCs.
Much faster than any CRC algorithm that doesn't use
a massive lookup table (which better not be bigger
than your L1 cache).

	David

-- 
David Laight: david@l8s.co.uk