RE: draft-ietf-v6ops-addcon-03.txt ... ULAs of shorter-than-/48 and ULA multicast scope matching ...

[Christian Huitema <huitema@windows.microsoft.com>]

On Friday, March 16, 2007 Tim Chown wrote:
>
> On Fri, Mar 16, 2007 at 03:04:17PM +0100, Brian E Carpenter wrote:
> >
> > On 2007-03-13 16:28, Tim Chown wrote:
> >
> > >Fair comment.   I guess we're also nibbling here at the 'what's a
> site'
> > >question.  I don't know where Thomas Narten's 3177-bis proposal got
> to,
> > >but if end sites were allocated a /56 you'd probably want a /56
> ULA...
> >
> > A /48 ULA would do just fine. They aren't going to be in short
> supply.
> 
> I would assume doing that would affect the collision probabilities?

Basic maths: the probability of at least one collision is proportional
to the number of allocations (networks) and inversely proportional to
the size of the address space. With a /48, the ULA address space is 40
bits wide, 2^40. If you run the numbers, the probabilities of at least
one collision for the various examples cited in the discussion are:

	For 256 independent /48 prefixes: 6 E-8
	For 2 independent /40 prefixes:   9 E-10
	For 1,000,000 independent /48 prefixes: 0.9 (90%)
	For 1,000,000 independent /56 prefixes: 0.003 (0.3%)
	For 1,000,000 independent /64 prefixes: 1.4 E-5 (i.e. a success
rate of 99.9986%, not quite five nines)

Of course, this can be mitigated in all sorts of ways, e.g. running a
registry...

-- Christian Huitema