Re: [homenet] Medius: additional DHCPv6 (PD) options needed?

[Ole Troan <[e-mail]>]

Ted, et al,

> I think the configuration model here needs more thought.   The hipnet document talks about heuristics for figuring out that it's at the edge, but e.g. the /48 heuristic seems like it won't work.
> 
> What I would like to see is something like Fred Baker's draft (draft-baker-homenet-prefix-assignment-01), for any router that could be a CER.   Fred's document has a few omissions that would need to be dealt with:
> 
> (1) It doesn't describe how DHCP relay works.   A non-CER relays to the CER, the IP address of which it got in the CER option.   If the non-CER knows of more than one CER, it relays to all CERs.
> (2) A router can be both CER and non-CER, in a multihoming situation, in which case it both answers and relays all DHCP messages.   This is slightly weird, but the DHCP protocol allows multiple servers to respond, so it should be okay.
> (3) routers that get multiple offers of different prefixes have to configure both prefixes, not just one, so that multihome prefixes propagate into the network.   This is not what RFC3315 and RFC 3633 advise; doing what they advise will result in a network that is multihomed at the edge, but not in the center.
> (4) routers that relay DHCP messages sniff for routes on the replies, and use them to populate their routing tables.
> (5) Fred's draft suggests that two routers advertising prefixes on the same link is an error, and that one of the routers needs to let the other win, but this misses the dual-home dual-CER case, where the prefixes being advertised are from different providers.   So the draft needs to be narrowed to the case where both prefixes come from the same provider, which probably means that prefixes need to be labeled as to the provider with a new DHCP option.
> (6) I'm probably missing some routing subtlety here.
> 
> This sounds complicated, but I think it's simpler than it sounds.   However, I think the current state of the CER and hipnet specs is _too_ simple, and will not actually work.   You have to solve the problem that Fred's draft solves, and Fred's draft is actually pretty close.

please read http://datatracker.ietf.org/doc/draft-arkko-homenet-prefix-assignment/
before we start building spanning trees with DHCP.

I'm in favour of hipnet as an intermediate step for a restricted topology. e.g. two routers in series,
with only one CER (that can potentially be multi-homed).

I'm not in favour of using hipnet for the general problem of an arbitrary topology.

border discovery is tricky. you can define some heuristics, but without some sort of security protocol it is going to be trial and error. in the eRouter case it is trivial of course.

cheers,
Ole