draft-gont-6man-slaac-renum (was: Re: [v6ops] cpe-slaac-renum: Proposed text for prefix lifetimes)

[Fernando Gont <[e-mail]>]

On 8/4/20 17:46, [e-mail] wrote:
>>> Sorry, I don't see the gap.
>>> All deployments I am aware of use the remote-id / option-82 equivalent option to identify subscribers.
>>> Most operators are a lot more clueful about this than us here.
>>>
>>> Think of PD as a contract. The ISP promises to lease the prefix to the subscriber for the time stated.
>>> The ISP can still do flash renumbering (although that's likely to induce a customer call) in two ways.
>>> One, just ignore it's earlier promise and give the subscriber a new prefix (which introduces all the problems Fernado wants to use prefix lifetimes to solve), or it can include the old prefix with lifetimes 0 (violating it's earlier promise, but at least being explicit) and advertise a new prefix.
>>
>> I’m trying to figure out why this is such an utterly unsatisfying response.  Yes, I get that in the ideal case, which is what you are describing here, there is no problem.  But you are missing _all_ of the edge cases.  And you’re assuming that things on a home network are a lot cleaner than I think is safe to assume.
>>
>> So yes, in the case you describe, where everything is done correctly, there is no problem.  I would not expect to see the persistent prefix problem routinely.  But I think it’s worth designing for—there are a lot of reasons why renumbering might happen in a way that is not as clean as you’re saying it should be, and we ought to handle it gracefully when it happens.
> 
> s/ideal/typical/
> 
> This small problem has a couple of simple solutions:
>   - change to an ISP that operate better

Seriously? Please think twice. (Hint: in many countries there's only one 
ISP -- welcome to the real world).



>   - the CPE sends back an ICMP in these cases, it's just for the host to act upon it

We devoted a section to this, such that we can save electrons in the 
emails: 
https://tools.ietf.org/html/draft-gont-6man-slaac-renum-06#appendix-B.1



> A host that has multiple addresses with different reachability is typical in IPv6. Every host is essentially multihomed. With ephemeral addresses.
> The case you describe above is not any different from that. The host-stack/application needs to react to the connection not making progress and try a different SA/DA pair.
> Similarly on initial connection.

We're fine with you boiling any ocean you want.



> You don't solve that problem by tweaking prefix lifetimes.

As Ted noted, getting rid of insanely long prefixes is part of that.

Basic stuff that can make a fresher prefix be used, and old addresses to 
be garbage collected.

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