Re: [v6ops] I-D Action: draft-gont-v6ops-ipv6-addressing-considerations-00.txt

[Fernando Gont <[e-mail]>]

On 5/1/21 13:08, Gert Doering wrote:
> Hi,
> 
> On Tue, Jan 05, 2021 at 02:24:14PM +0100, Eduard Metz wrote:
>> Maybe the last column should be "Applicable to Internet" (no / no / yes)
>> or something along these lines. I assume the point here is to indicate that
>> link-local and ULA cannot be used on the Internet.
> 
> What is "the Internet"?
> 
> We're an ISP, so if we use ULAs on our infrastructure, outside any
> firewalls (but not announcing to other ASes, or leaking packets) - is
> that "used on the Internet" or not?

Agreed that "Applicable to Internet" is probably not a good taxonomy.

The proper (and useful) taxonomy is that of "scope". ULAs clearly don't 
have a global scope (as per the definition in RFC4007).

Given a ULA prefix (say, fd12:3456:7890:abcd::/64), it doesn't globally 
(and uniquely) identify a link. Similarly, nobody is in the position of 
publishing a ROA or restricting which AS can originate packets from such 
prefix.

IMO, generating the "Global ID" of ULAs from a PRNG does reduce the 
possibility of collisions *when a limited number of ULA-based networks 
are interconnected* -- but the birthday paradox tells you can't expect 
the ULA prefixes to be globally unique. There's more to "global scope" 
than simply asserting that if you interconnect, say, two networks the 
chances of colliding prefixes will be small.

Thanks,
-- 
Fernando Gont
SI6 Networks
e-mail: [e-mail]
PGP Fingerprint: 6666 31C6 D484 63B2 8FB1 E3C4 AE25 0D55 1D4E 7492