[IPv6]Re: AD review for draft-ietf-6man-slaac-renum-13

[Lorenzo Colitti <[e-mail]>]

On Sun, Jul 5, 2026 at 3:32 PM Fernando Gont <[e-mail]> wrote:

> FWIW, I did the obvious: recommended a default of 12, that corresponds
> to the ND_RAS_VALID parameter in the same expression that we use to
> compute the ND_DEFAULT_VALID_LIFETIME, and thus results in a probability
> of the stale information being discarded of 0.99975.
>
> (i.e., we no longer recommend stale information to be advertised for a
> certain period of time, but rather that it is transmitted a certain
> number of times).
>
> Please do let us know if this addresses your concern.
>

I don't think that's sufficient because it does not state how frequently
those RAs need to be transmitted.

For example, a router could implement the current text as follows: "in the
next ND_RAS_VALID RAs that are sent, include the old prefix with a valid
lifetime of zero". This will indeed send 12 advertisements, but it will
send them over the course of (by default) 1 hour. That does not address my
concern because it means that devices that drop RAs will stay broken for a
long time. It also does not address my previous concern that if the
configuration changes frequently, then the RA will become filled with old
prefixes to be deprecated.

I think we need these zero-lifetime RAs to be sent out "more frequently,
soon after the renumbering happens". I'm not sure what "more frequently"
and "soon" should concretely mean. A start might be to say that the
behaviour should be the same as when an interface becomes an advertising
interface. That wouldn't be particularly fast,
since MAX_INITIAL_RTR_ADVERT_INTERVAL is 16 seconds, so it will result in
the 12 RAs being sent over a period of 192 seconds (just over 3 minutes),
but that might be fine.

BTW: how strongly do we feel that the valid lifetime should be reduced from
5400 seconds (RFC 9096) to 3600 seconds? The document would be simpler if
we didn't change 5400 and instead just set ND_RAS_VALID to 18 instead of
12. But I don't remember the history here.