Re: [v6ops] draft-ietf-v6ops-host-addr-availability WGLC

[Mark Smith <[e-mail]>]

On 22 Sep 2015 11:22 pm, "Mikael Abrahamsson" <[e-mail]> wrote:
>
> On Tue, 22 Sep 2015, Mark Smith wrote:
>
>> I've never been completely sure whether IPv6 supports the weak or strong
host model, or whether it is possible for a host/implementation to be one
or the other.
>
>
> I'd say a bit of both.
>
> I find that systems that will respond to an arp query for its eth1
address and the arp request comes in on eth0 to be fundamentally broken.
Some people do not agree. From this aspect, I'd say my take is that IPv6
uses the strong hos model.
>

I agree, and I think it is probably because I first got exposed to
multi-address issues on routers, and they followed the strong "host" model.
Addresses are assigned to interfaces, and if the interface goes down or is
taken down, the associated addresses become unavailable. I find that very
intuitive because the address state follows the interface state.

> However, most hosts will glady send out traffic sourced from eth0 IPv6
address via eth1, if that happens to be the best default route for the
node, indicating weak host model.
>
> With hosts implementing MIF, I hope most things will be strong host model
only.
>

OTOH, if an interface suffers a transient fault, should all of the sockets
that are using the interface's addresses be immediately destroyed? I'd say
no, but that then means the weak host model is the better and more robust
one.

Actually, this has reminded me of probably what I'd prefer on hosts. Weak
host model by default, meaning addresses and associated sockets survive
transient link faults, strong host model if interfaces are administratively
taken down, so addresses and associated sockets disappear when intended by
an operator or process.

I think this starts to uncover two different behaviours behind the strong
and weak host models:

- is an address still available for a host to use/listen to if the
interface the address is associated with goes down and/or is taken down
administratively?

- if the address is still available when the associated interface is down,
should packets using that address be routed out other interfaces, or should
they be dropped.

Regards,
Mark.

>
> --
> Mikael Abrahamsson    email: [e-mail]