[arin-ppml] What is a "host"?

On Mon, Sep 20, 2010 at 11:53 AM,  <michael.dillon at> wrote:
>> So, we really have to go back to basics if we want to do something
>> sensible with v6 policy. There are two foundations for PI addresses:
> Then count interfaces.
> In IPv6, addresses are assigned to interfaces, and networks are composed
> of subnets, which contain a bunch of interfaces.

Six of one, half a dozen of the other.

Go back to basics. The foundations are about dollar cost. In fact,
they're about two specific dollar costs.

The worldwide systemic cost of introducing an IPv6 route into the BGP
table is somewhere between $10k and $15k per year by now.

Renumbering has a cost as well, based on the manpower spent changing
configurations, debugging the changes that don't work, and the cost of
the inevitable service disruptions during the renumbering process.

In the multihoming case, the BGP cost is the same either way, so the
renumbering cost is always higher, except to whatever degree folks
would decide to add multihoming to meet policy versus multihoming to
achieve required system resilience.

In the singlehomed case, the questions are:

a. How much more than the BGP cost should the renumbering cost be
before it's reasonable to allow the BGP cost?
b. What metric do you use to assess the renumbering cost?

> Given that we are talking about IPv6 renumbering, not IPv4 renumbering,
> subnets still seems like a good measure. In IPv6, the subnets are unchanged
> and you just change the network prefix handed out by SLAAC or DHCPv6.

Respectfully, the assumption that IPv6 renumbering will prove
meaningfully less costly than IPv4 renumbering is still unfounded.
Wait for it to play out in large, real-world deployments before
building this assumption into policy.

Bill Herrin

William D. Herrin ................ herrin at  bill at
3005 Crane Dr. ...................... Web: <>
Falls Church, VA 22042-3004