[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: new draft on IPv6 CPE router available for review



On Thu, 3 Jul 2008, Shin Miyakawa wrote:

(A)A machine will follow RFC3484 to select source address
  regardless if this machine is a strong host model or a weak host model
(B)No routing function (for example, because there is no LAN behind)

Correct.

(X)Even if a machine terminates PPPoE, L2TP or whatever uplink by itself
  with no other machines in the customer premises, this machine must have BOTH models.

I never said all IPv6 speaking HOSTS need to do this. For me a CPE is something that is involved in the delivery of a service, not an end user host device. I see two different services, one which is for connecting a single device and nothing else, and another service where the customer has a router and several hosts behind it. If this is another definition of a CPE, then I need to change my terminology to avoid confusion. Am I using the wrong terminology?

For example, if we create sensor networks, do we still need routing function on all the sensor devices ? I do not think so.

Of course not.

At the same time, please note that we have no way to identify a CPE is a router
or just a host.

The customer will have to decide this at the time of purchase of the service.

Of course, "The Service MUST support BOTH models" will be fine.
           ===========

That is not what I want, quite the opposite.

So, let me elaborate my view, and include my definitions which might make it more clear what I'm after.

For me, a CPE is the device that connects the home network to the outside world, it might be a DSL modem, a NAT gateway, wireless access point or alike. It is NOT an end host.

In 10 years time, I see people having all kinds of devices in their home, and the ISP does not want to keep state or forwarding resources for every device, thus we need to route IPv6 subnet to a router that is placed in the home, controlled by the customer (or a provisioning system in case of a managed CPE, but the customer will most likely be influencing how the CPE is configured thru a selfcare system that then configures the CPE).

For people with only a single computer in their home, I want this to be a special case, and since they don't need triple play or alike, we might as well tunnel them over IPv4/ethernet in the interim, with 6to4, PPPoE or alike. So the ISP provides two different services whether the customer wants to have a home gateway with a LAN behind it, or if they want to just connect a single computer.

So, my rationale is that any home gateway device needs to support both models (link-local only for WAN and the strong host model), end hosts only need to support the strong host model.

Does this make my position more understandable?

--
Mikael Abrahamsson    email: swmike@swm.pp.se