Re: A common problem with SLAAC in "renumbering" scenarios

[Jan Zorz - Go6 <[e-mail]>]

On 13/02/2019 16:38, Fernando Gont wrote:
>> It could also just include it as a hint in the first SOLICIT.
>>
>> Storing the old lease to persistent storage also gives the CPE
>> visibility if the prefix does change.  If it does, it can then send
>> RAs with lifetime 0 to deprecate the old prefix on the LAN side.  This
>> seems to have resolve all of the issues for us, as far as we can tell.
> 
> It may solve the issues when/if implemented. For obvious reasons, you
> cannot rely on this functionality being deployed -- hence it does nto
> relieve us from improving the robustness of hosts.

Correct, I see the benefit in making host more robust and resilient - 
and making this a standard can only and just improve things, can't make 
them any worse.

However, CPE is also an important part of the game here. As said before 
and as heard from Richard, some huge operators are already deploying 
this "hack" where they write a PD to persistent storage and compare 
prefixes after reboots and this solves a lot of problems. Why not making 
this easy fix a recommended thing to do for everyone?

When talking about the provisioning part - IETF can't tell operators how 
to run networks (obviously), but what we can do here is to say "when 
assigning IPv6 prefix to a customer, identified by the same DUID, we 
must not assign a different prefix to the same DUID before the lifetime 
of PD that we set doesn't expire, otherwise you are breaking the 
protocol standards."

If somebody decides (for some bizzare reason) that it's a good idea to 
make CPEs with randomized DUID after every reboot, then they shall 
implement memorizing PDs in persistent storage across reboots and send 
ICMP packets after reboot to deprecate old prefix(es)

I think that extending a scope of our I-D to this three "advices" would 
bring at least some hope for improvement, as currently it can't be 
worse. Of course, things are easy for huge operators with home-brew CPEs 
and a lot of knowledge, but there are other, smaller and less deep-tech 
educated ISPs that could benefit from improved defaults.

Cheers, Jan