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

[Fernando Gont <[e-mail]>]

Ole,

On 17/2/19 17:23, Ole Troan wrote:
[....]
>> 
>> Agreed, Tom. But that only strengthens the point that going to
>> great lengths to make IPv6 prefixes stable, when we've spent so
>> much time doing the opposite with respect to the IID, seems
>> self-contradictory. It seems more productive to fix whatever
>> breaks, when the prefix changes. For instance, Fernando's points
>> about SLAAC timers. That should be the main focus of the IETF, IMO.
>> Ensuring permanent prefixes should be the odd exception, done with
>> more or less extraordinary measures.
> 
> Lifetimes become quite useless, when the promise made with them is
> broken. They only serve then for garbage collection.

This is tricky. Robustness has a lot to do with being able to gracefully
deal with situations that "shouldn't happen".

If you asked me, yes, I'd love the network to behave nicer. That said, I
also want the network to be robust, and not break when e.g. a
crash-and-reboot event takes place.



> If your model of networking is dynamically changing addresses, then
> don’t you need a rendezvous point for all inbound communication
> anyway? Guess then you will not have much benefit from a 128 bit
> address space either.

I think you are mixing things up here.

1) No matter whether dynamic or stable prefixes are in use, the network
should be robust. The discussion of stable vs dynamic prefixes is
orthogonal to this document.

2) Many have argued in favor of temporary-only which, in some deployment
scenarios (e.g. mobile), makes sense. In scenarios where you'd go
temporary-only, then dynamic prefixes obviously make sense. Folks have
also noted reasons for which they may want to do dynamic prefixes. Your
network, your policy.

3) With "only allow outgoing communications" (RFC6092) plus IPv6 CPEs
not supporting uPNP, you do need a rendezvous point. Ironically enough,
I have dual-stacked boxes which I can reach directly via IPv4 (DDNS +
UPnP) but not over IPv6 (no way to punch holes in the v6 firewall). From
this point of view, the justification for v6 is that when/if we
eventually get to multiple layers of NAT, UPnP will not do the trick,
and by that time the CPE vendors will have figured out they should
include UPnP in their IPv6 CPEs.


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