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

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

On 02/02/2019 18:19, Michael Richardson wrote:
> 
> Jan Zorz - Go6 <[e-mail]> wrote:
>      > On 31/01/2019 13:11, Fernando Gont wrote:
>      >>> Doesn't RFC7084 already say this? L-13.
>      >>
>      >> Yes, and we missed this (thanks!). -- that said, we added this bullet
>      >> for completeness sake. The case we care about in this doc is the
>      >> reboot scenario.
> 
>      > However, RFC7084 is Informational and not Standard. I think this should
> 
> Additionally, when it comes to what the CPE can expect from the ISP, 7084 is
> essentially a set of heuristics, and there is no document about what the
> *ISP* can expect from the CPE.

Hey,

I think that this is partially described in RIPE-690 BCP document.

> 
> We need to turn both in standards, with a clear palette of behaviours.
> 
> For instance, I'd like to see standardized support for:
> Scenario A:
>     a) do *not* number the PPP(oE) link in IPv6. (use LL only)
>     b) ISP delegates /60 or bigger and the CPE uses an address out of it.
> 
> Scenario B:
>     a) do *not* number the PPP(oE) link in IPv6. (use LL only)
>     b) ISP delegates exactly a /64, and the CPE uses an address out of it,
>        and uses the rest for tethering, single link.
> 
> Scenario C:
>     a) number the PPPoE link with RA and/or DHCPv6.
>     b) Use rfc6603 to clearly exclude the link prefix
> 
> Scenario D:
>     a) number the link with a /64
>     b) provide another /60+ if the CPE asks for it later on.

Maybe we can combine the wisdom agreed by operators in RIPE-690 and 
specify it in more detailed way, as you propose above.

> 
> For ISPs, there is a major win to having the all of the customer prefixes in
> a single prefix (routing entry), but it's hard to know which scenario is
> going to occur.

This one is the most optimal possible, but now we need CPE vendors to 
start supporting it.

> 
> There is not enough in RFC7084 for an ISP to provide requirements to PPPoE
> termination vendors (DSLAM) or to CPE vendors.    In the work I did at
> Finepoint, I basically had to invert 7084 to make my requirements, but I
> couldn't be sure what CPEs would support which scenarios.
> 
> The Broadband Forum TR-187 is useful, but still has too much wiggle room.
> 
> I would be willing to co-author a document.
> 

Excellent!!! Please, read RIPE-690 and see if we can get something out 
of there as a start ;)

Cheers, Jan