Re: [homenet] I-D.ietf-homenet-prefix-assignment (RFC 4193 conformance)

[James Woodyatt <[e-mail]>]

On Thu, Oct 9, 2014 at 2:04 AM, Pierre Pfister <[e-mail]>
wrote:

>
> What do you think of the following proposal ?
> It allows any router to generate a ULA (it adds more complexity because
> collisions must be avoided, even though the Backoff was necessary at boot
> anyway).
>

As this is a Standards Track requirements draft, I hope to be excused if
I'm seeming to be overly pedantic, but this revision still leaves me
concerned that we are setting requirements that do not admit Thread
networks into HOMENET routing domains.  Here is the excerpt that I find
most troubling:

ULA prefixes may be provided by DHCPv6-PD or static configuration, as
specified in Section 4.3, in which case they are not considered as
"spontaneously" generated and MUST NOT be withdrawn if another ULA
delegated prefix is observed.


My problem is that Thread networks autonomously number themselves with a
ULA /64 prefix, i.e. a ULA /48 prefix with a well-known 16-bit subnet
identifier. This happens when the first element in a disconnected network
commissions itself. Thread networks will keep this /64 prefix as more
elements are commissioned with network credentials to join the mesh. In the
highly likely event that one or more elements capable of acting as Thread
border routers to the HOMENET domain later join the mesh, then each of them
would naturally want to advertise into the HOMENET domain this autonomously
generated ULA /64 prefix. The language in this proposed revision still
seems to admit only ULA delegated prefixes obtained by DHCPv6-PD and by
*static* configuration, but neither of those are applicable in the case of
Thread networks.

I think my concern might be ameliorated by drawing a distinction in the
requirements between a single distinguished Home ULA Prefix and any number
of other Exterior ULA Prefix delegations.  The former prefix is
autonomously generated by the HOMENET router in the Leader role, whereas
the latter are delegated by exterior numbering authorities outside the
HOMENET domain, and they are just like any other globally scoped IPv6
network prefix.

Would it be helpful if I tried to write up a proposed set of edits for this?


-- 
james woodyatt <[e-mail]>
Nest Labs, Communications Engineering