Re: Adoption Call for "Improving the Robustness of Stateless Address Autoconfiguration (SLAAC) to Flash Renumbering Events"

[Mikael Abrahamsson <[e-mail]>]

On Fri, 26 Jun 2020, Bob Hinden wrote:

> We would like the w.g. to consider and comment on these issues when 
> responding to this adoption call.

I agree that there are problems to be solved in this space (thus I support 
adoption), but I think that the current version of the draft is way too 
overengineered and "heavy handed". I think we can achieve all the benefit 
without changing so much of current behavior.

For new communication, hosts will primarily use the GUA address with the 
highest preferred value. So this is where we should focus the effort. 
Having addresses based on valid-lifetimes around that are no longer 
functional but are not used is of very little real life problem in my 
experience. It's mostly a cosmetic problem.

In the topology of:

RGW <-> L2-switch <-> host

We can easily get RGW changes without link-down to the host. This is the 
same for wifi case if there are separate APs from the RGW. So there are 
problems that can occur without host L1 down as a signal that something 
changed.

If the RGW is power cycled and comes up with lower timers (perhaps because 
new ISP or old ISP changed DHCPv6-PD timers) we might now be in a case 
where the old PIO based addresses might have higher preferred value than 
the new RA/PIO that the RGW is now sending. So there is a problem to solve 
here.

However, I do not agree with all the valid-lifetime changes and arbitrary 
cap:ing (where did the value "48 *" come from?). I see no major upside in 
restricting network manager choice when it comes to valid times.

So perhaps we can come up with / keep some simple heuristics regarding 
lack of old PIO information and then at that point, we cap those addresses 
preferred values to be less than the currently seen PIO by the same 
router, or if there is another router announcing RAs. There might be a MAY 
section talking about connection checker kicking off for some of these 
cases to verify connectivity of those potentially no longer working 
addresses? I don't want to go too deep into solution space.

Just to iterate, I support adoption but I think the draft will need quite 
a lot of changes before I would support publication.

-- 
Mikael Abrahamsson    email: [e-mail]