RE: RFC7084

["STARK, BARBARA H" <[e-mail]>]

> Do these networks also support hosts plugging in directly?
> Designing a network where DHCPv6 is a MUST would seem to me to be a violation of the stated node requirements, where it is merely a SHOULD.

First I'll tackle the 2nd sentence. The node requirements (RFC6434 -- an informational RFC) define the most basic capabilities of what should be expected of an IPv6 node. There is no requirement anywhere that all networks must be designed to allow all nodes (that meet these basic requirements) to connect to them. It is also (IMO) a well-established practice that it's not ok to weaken a referenced RFC's requirements (changing MUST to SHOULD), but it is ok to strengthen them (SHOULD to MUST) when defining a specific subset of devices that are expected to go above and beyond that RFC's expectations. It's also (IMO) perfectly ok to define networks that only allow for connectivity by such a subset. The decision on whether or not to do that is generally based on whether it's possible to reasonably expect (or set expectations) that users of the network will understand and accept such limitations on the set of connectible hosts, and whether there's benefit or good reason to limit the set.

My approach in designing CE routers is that the LAN side of the router needs to allow for connectivity by the most basic RFC 6434 nodes. The RFC7084 LAN-side requirements are intended to accommodate this. That is because consumers (reasonably) expect that all their smart TVs, smartphones, PCs, tablets, NAS boxes, etc. will "just work" when they plug them in. The LAN side of the CE router needs to meet this expectation, or there will be trouble calls. 

Consumers (as a generalization) do not expect to be able to connect their smart TVs, tablets, etc. directly to any access network and have it just work. Very rare is the access network where this works. Therefore, this is a well-set expectation limit and service providers consider it ok for access networks to have tighter expectations and not support connectivity by any RFC6434 node. This does not result in trouble calls. In fact, by having tighter expectations, trouble calls are reduced. When the company I worked for 10 years ago started shipping really basic DSL modem/routers to all customers and stopped offering DSL PCI cards, DSL USB modems, PPPoE PC clients or bridged DSL modems, trouble calls with installation went down dramatically. 

So now I'll answer the first question. Instead of interpreting the question literally (yes, a RFC7084 CE router is also a host and can plug in directly, therefore there exist hosts that can plug in directly), I'm going to interpret the question as "Do these networks also support *all RFC6434-compliant* hosts plugging in directly?" 
No. They do not. I can't speak to the CableLabs-defined access networks, but BBF TR-177 access networks are only defined for connectivity by DHCPv6-capable hosts.
From TR-177:
"The main focus of TR-177 is on following scenarios:
 - a subscriber having a routed RG. This RG can be connected to the access loop either directly
or through a bridged RG. The routed RG acts as a Requesting Router (as per RFC 3633), and
possibly embeds DHCPv6-based address allocation capabilities.
 - a subscriber having a bridged RG, in which case the first hop router for their hosts is the
BNG. The subscriber hosts support DHCPv6 address allocation capabilities."

Some service provider instantiations of TR-177 access networks probably won't even support all RFC7084-compliant CE routers, because those service providers require customers to use the provider-provided routers. For those who are upset by this statement: Please don't see my saying this as my having any sort of emotional attachment to this practice or recommending it. I'm just saying it because it is an accurate statement that may be useful to understand. It is what it is.
Barbara