Re: [v6ops] WGLC: draft-ietf-v6ops-unique-ipv6-prefix-per-host-02 - multiple prefixes per device

[Lorenzo Colitti <[e-mail]>]

On Fri, Mar 17, 2017 at 2:31 AM, Brian E Carpenter
<[e-mail]> wrote:
> Yes, but only if /64 is no longer sacred. It's going to be tricky if some
> of the nodes on the /66 subnets assume SLAAC at /64.

If /64 is no longer fixed then we'll end up with /128 + NAT66 like we
do in IPv4.

> So, it seems clear to me that while getting a /64 is better than getting
> a /128, it's even better to get a /56. Or even one of those 15 trillion
/48s.
> We should make it clear that /64 is not even second-best. Just better
> than /128.

Given that this is an operational draft, then I would suggest an
applicability statement: state that DHCPv6 PD is better, but also state
that best *current* practice is to provide a /64 via RA, because currently
very few hosts have DHCPv6 PD clients and providing PD on the network side
is not going to help the vast majority of clients.

Using shared links on public deployments with untrusted hosts has all sorts
of security and scalability issues (ND spoofing, DAD spoofing, ND cache
exhaustion attacks, etc.) /64 to the host is just a much better solution.