Re: Non routable IPv6 registry proposal

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

David,
On 12-Mar-21 08:19, David Farmer wrote:
> On Thu, Mar 11, 2021 at 10:49 AM Keith Moore <[e-mail] <mailto:[e-mail]>> wrote:
> 
>     On 3/11/21 5:22 AM, Nico Schottelius wrote:
> 
>>>     Another question I have is whether such ULA allocations
>>>     will realistically remain local.
>>     ULAs are unlikely staying local, as we have seen with radio networks in
>>     Germany. Tunnels are being used to interconnect remote cities and
>>     non-collision (not necessarily public routing) are a primary concern.
> 
>     Despite the name, there's no reason that ULAs should stay local.   As long as they are properly chosen, it's perfectly reasonable to route them privately between cooperating networks, and IMO this is part of their design.   One of the problems with RFC 1918 addresses in IPv4 was that enterprises had a need to route traffic between networks each using that space.   The resulting address collisions generally required explicit NAT configurations to work around, and these were failure-prone and difficult to manage.  ULAs were intended in part to remedy this problem.
> 
>     Keith
> 
> The "L" for Local isn't intended to have a strict definition of Local. However, similarly, the "U" for Unique isn't intended to have a strict definition of Unique either, especially a mathematical definition of Unique. 
> 
> You can easily interconnect thousands or even tens of thousands of ULA prefixes without much chance of an address collision, as long as the random assignment process is actually used. Whereas, if you try to interconnect billions of ULA prefixes, you will probably start running into the birthday paradox.
> 
> So the interconnection of ULA prefixes, the route-ability of them, is not intended to be unlimited. There are limits to the number of ULA prefixes that SHOULD be interconnected to each other; nevertheless, this limit is extremely generous for the intended use cases.
> 
> If you disregard the intended use cases and use them outside the intended use cases, then address collisions could become an issue.

I'm not sure where you get your "intended" from. I don't think we've ever really written done the intended use cases in such detail. (Except for the abandoned https://datatracker.ietf.org/doc/html/draft-ietf-v6ops-ula-usage-considerations-02 )

    Brian