On Tue, Oct 14, 2014 at 12:31 PM, Ted Lemon <[email protected]> wrote:

> [...]
> This is where I am just completely puzzled.   We talked about this
> previously.   I thought the idea was that the homenet ULA should converge:
> that there should only be one, ultimately [...]


This is exactly what I'm trying to surface in my earlier comments about
I-D.ietf-homenet-prefix-delegation. That idea needs clarification if we're
going to interoperate with network layers like Thread which have their ULA
prefix that it would be good to advertise in HOMENET domains as a delegated
prefix.

If the idea is to minimize the need for a HOMENET autonomously generated
ULA prefix, then it should only be advertised when not other ULA prefix is
available and it should be deprecated and allowed to expire when it isn't
needed.  If on the other hand, we do not see a need to limit the number of
ULA prefixes advertised into the HOMENET domain, then a persistent one
should be generated when the network is commissioned by its first leader,
and it should always be advertised thereafter whether a first-mile service
is operational or not, and regardless whether the initiating leader leaves
the network.  (There is a problem with the latter case, which is that some
legacy host operating systems are still broken in an environment like that,
and it would be helpful to mitigate such brokenness. The former case
doesn't have that problem.  There is also the exception that arises when
two networks with different ULA prefixes are joined— now you have one
network, with two ULA prefixes, neither of which can ever be allowed to
expire.)

On Tue, Oct 14, 2014 at 12:44 PM, Brian E Carpenter <
[email protected]> wrote:

>
> At a stroke this would destroy the main advantage of ULAs -
> namely, invariant addresses for internal traffic. IPv6 assumes
> multiple simultaneous addresses; there is no reason whatever to
> artificially prevent use of ULAs alongside GUAs.
>

p1. I don't want to prevent the use of ULAs alongside GUAs. Indeed, I need
for this to be preserved, and I'm very concerned about requirement language
that would seem to interfere with that.

p2. While I'm in agreement there is a benefit in a guarantee for hosts on
home networks that they will always have valid addresses in the interior
routing domain, I'm not sure I can agree that the main reason to use a ULA
prefix is to encourage the supposition that a HOMENET generated ULA is more
stable and persistent than any GUA assigned by a first-mile service. I
suppose if the working group has already argued that to death, and
concluded that stable persistent addressing solves a problem that real
people are actually facing, then it's not worth rehashing that discussion.

If we're going to go with HOMENET always generating a ULA prefix at network
commissioning time and persisting for the life of the network, then I'm
going to need to understand better how we're handling network joins and
splits.


-- 
james woodyatt <[email protected]>
Nest Labs, Communications Engineering
_______________________________________________
homenet mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/homenet

Reply via email to