Re: [dane] hash truncated to 28 octets

[Viktor Dukhovni <[e-mail]>]

On Tue, Aug 04, 2015 at 12:12:45PM +0000, Viktor Dukhovni wrote:

> On Tue, Aug 04, 2015 at 03:50:32AM -0400, Paul Wouters wrote:
> 
> > >2,since some local-parts are longer than 28 octets, are there some collisions after hash?truncated?to?28?octets??
> > 
> > I think if you have 100.000 email addresses in one domain, the chance of
> > collision would be pretty small. but non-zero.
> 
> You'd need 2^112 email addresses for an appreciable chance of
> collision.  If every person on the planet (say 10^{10} people some
> day) each had 100 email addresses in the same domain, that'd be
> 10^{12} or ~2^{40} addresses.  The collision probability would be
> around 2^{80}/2^{112} = 2^{-32} ~ 10^{-6}.  With "just" 10^{10}
> addresses, it drops by a factor of 10^4 to 10^{-10}.

For the record, the above vastly overestimates the collision
probability.  I accidentally computed the probability for a 112-bit
hash, not a 224-bit hash.

For a 224-bit hash, the collision probability with 10^{12} addresses
is around 2^{80}/2^{224} or 2^{-144}.  No collisions are likely
before Earth is incinerated by a red-giant Sun around 2^{85}
nanoseconds from now.

-- 
	Viktor.