Re: Happy St Nicholas Day: Re-Launching the IPv6 ULA registry

[Brian E Carpenter <[e-mail]>]

On 07-Dec-20 21:28, Fernando Gont wrote:
> Hi, Lorenzo,
> 
> On 6/12/20 21:42, Lorenzo Colitti wrote:
>> Nico,
>>
>> This seems a bit misguided for several reasons.
>>
>> First, registries are supposed to be authoritative databases, but the 
>> RFC specifies that ULA addresses are generated pseudo-randomly.
> 
> Strictly speaking, there exist registries that are not authoritative -- 
> e.g., think of the registered ports in the port number registry.

That's not a useful analogy. Those ports need to be registered because
they are potentially used across the open Internet. ULAs are by
definition not used across the open Internet. Therefore, they don't
need a registry.

(That's why we've never reached any kind of consensus about ULA-C:
by definition, they don't need to exist.)

(Nico: but thanks anyway. I was a bit sorry when that aspect of sixxs
went away.)

>> Therefore, if a pseudo-randomly allocated address conflicts with 
>> something in the registry, the registry should be deemed incorrect.

Not even incorrect (to echo "not even wrong" [1]).
 
> Following the logic of the port number registry, if two organizations 
> use the same prefix, the registry could simply note that.

It could. As Lorenzo hinted, the most frequent entry would be zero.

Any arbitrary string of digits occurs an infinite number of times
in the expansion of π. That means that your ULA prefix also occurs
an infinite number of times in the (hexadecimal) expansion of π.
That fact is as much use as a ULA registry (a list of random numbers).

> That said, I'm not sure:
> 
> 1) Whether folks would be compelled to register their blocks.

I'm sure: no.
 
> 2) (Partly as a result of #1), the extent to which a registry would be 
> of any use.

I'm sure: no use at all.

> That aside, in a way the whole point of randomizing the ULA prefix is 
> that results statistically unique. So, provided folks do follow the 
> advice in the ULA spec (I for one have *not :-) in a number of 
> occasions), there shouldn't be much of a need to actually register ULA 
> prefixes... (if the prefixes are unique... who cares who uses what?)

Sorry, but I don't think "statistically unique" is a well-defined term.
The property of a ULA prefix is that there is a very low probability
of collision if two ULA networks are merged. (For the details, see
https://tools.ietf.org/html/rfc4193#section-3.2.3).

If I was writing a script to merge two networks I'd have to test for
that case, but despite the birthday paradox, I don't think I'd put
more effort into it than throwing an exception.

BTW, what I found useful in the old sixxs tool was the ULA prefix
generator. Not sure why ungleich wants me to log in to use that feature.

Regards
   Brian

[1] https://en.wikipedia.org/wiki/Not_even_wrong