On Fri, Jul 29, 2005 at 12:18:30 -0400,
"John D. Burger" <john@mitre.org> wrote:
> >If you use a large enough space for the number you can reduce that
> >probability of an accidental collision to much less than that of
> >catastrophic hardware failure at which point it isn't noticably better
> >than having no chance of collisions.
>
> I find the comparison unconvincing - if my hardware crashes, I know it
> and can decide how to recover. If two UIDs collide, my system may
> silently do something that may never be detected.
If it crashes yes, if a bit flips maybe not.
Note that by using a larger hash and more random bits you can make this
probability arbitrarily small. For 512 hashes with 512 bits of entropy,
I doubt you could compare documents fast enough to have a 50-50 chance
of finding a collision before the heat death of the universe.