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

Re: Discussion of the Home/SOHO environment



On Jan 3, 2008, at 15:22, Fred Baker wrote:

I would like the v6ops community to discuss this and come to some conclusion that can guide the CPE Requirements development. It may be worthwhile documenting that conclusion and the assumptions it is based on in an RFC. We are dealing at least in part with a world view question, and we need a rational and agreed world view.


Mr. Durand suggests that IPv6 presents service providers with an opportunity to offer a new, revised social contract to retail ISP customers, i.e. a very different one from the traditional social contract that IPv4 service providers have been offering, and with which retail customers have grown accustomed.

On January 3, 2008 at 1:22:01 PM PST, Alain Durand <alain_durand@cable.comcast.com> wrote:

In IPv4 broadband land, there is a pretty well accepted "social contract":

   - Customer get one IPv4 address that can change over time
   - Customer use/rent/own a NAT box to create more address space,
     isolate himself/herself from external IP address change, get
the so-called security benefits of NAT or whatever over local reason - the "security model" is mainly defined as: all devices within the home
     network belong to the customer, are mostly unmanaged and
a security perimeter is defined by the home router to "protect" the
     good inside from the "evil" outside.
- The ISP has very little if any view of the devices in the home south
     of the home gateway
   - There is little DNS in place

Note: all this is a direct consequence of the NAT model

In the brave new world of IPv6, the plethora of address space impose on us to revisit this model, mainly because NAT is not required to connect more than one device. Note that I said not required, which does not mean it will not be part of the picture in one form or another, if only in the NAT v4/v6/v4 that I described last IETF.

I couldn't agree more, but as one of those players with "skin" in the retail CPE development game, I'd like to point out that our customers are highly resistant to learning the difference between IPv4 and IPv6. They don't want to know, and they really don't see why they should have to care. Frankly, neither do I— on both counts.

The new IPv6 social contract needs to be functionally equivalent to the familiar old IPv4 one, or IPv6 service will have to be sold separately and marketed as an *alternative* to IPv4, not a complement to it. I'm pretty sure no one wants that (but I'll allow that I could be insufficiently cynical).

If we're going to revisit the IPv4 social contract for the brave new world of IPv6 broadband retail service, here's what I think it will have to look like to keep from scaring our customers:

...beginning with the proposition that what they already know isn't going anywhere...

+ Customer gets one IPv4 address, possibly from RFC 1918 space, that may change over time, sometimes several times per day. (I know of one service that changes my address six or seven times while I'm in transit between Cupertino and San Francisco.) + Customer uses/rents/owns an IPv4 NAT-gateway with all the traditional behaviors and idiosyncracies. + The ISP has very little, if any, view of the devices in the private IPv4 interior subnet south of the NAT-gateway. + The ISP offers very little, if any, DNS content service on behalf of the customer for their one IPv4 address.

...in addition...

+ Customer gets one global, provider-aggregated, unicast IPv6 prefix. (We can argue about the size. I have reasons for wanting / 48 over /56. I'm willing to haggle.) + Customer prefix may change over time, preferably with new prefixes added while the old one is deprecated and still usable for some additional time (measured in hours and maybe days). + Interior routing in the customer premises, if any, will be controlled by the customer, not the service provider. Zero configuration ad-hoc routing is how this will happen, if it does at all. + The IPv4 NAT-gateway owned/used/rented by the customer also serves as the IPv6 default gateway for their one IPv6 prefix. + On the customer interior network, the IPv6 gateway advertises its default route with RA. It may optionally offer nodes additional configuration parameters, under the administrative control of the customer, using DHCP6. Nodes without DHCP6 clients still get service by stateless autoconfiguration. + The ISP has very little, if any, view of the devices in the interior IPv6 network south of the gateway. (Our customers are downright truculent about this point. It's not a negotiable position. If IPv6 means changing this on them, then IPv6 is DOA.)

...finally, and this is IMPORTANT...

+ Customers with public IPv4 addresses get satisfactory 6to4 relay service.
+ Customers with RFC 1918 addresses get satisfactory Teredo service.

(Why are these transition mechanisms important? Because there's already a lot of installed gear that only supports IPv6 through one or both of these tunneling methods. When IPv6 addresses start flying around in application protocols, and devices using this existing gear start trying to communicate with nodes that have native IPv6 service, everything needs to work whether it's tunneled or not. Routing all my clients' IPv6 traffic (native from their perspective) through Ouagadougou because you don't think 6to4 service is important enough to treat with the same respect you treat your own native IPv6 transit is a great way to make me turn off IPv6 on all my servers and force all your IPv6-only customers to use the NAT-PT box you're still refusing to deploy.)

All the other questions Mr. Durand asks need to have the same answers for IPv6 as they currently do for IPv4.

Q: What is the management model of the home?
A: Zero configuration everywhere. Fire and forget provisioning. Plug in the router, authorize it once, and never mess with it again. The ISP can only usefully help by assisting with troubleshooting when connectivity fails.

Q: What is the "security model" of the home?
A: All devices within the interior network are in the customer's possession, mostly unmanaged, and the perimeter is defined by the gateway to "protect" the good inside from the "evil" outside. (Customer expectations about how this works for IPv4/NAT need to be respected when considering how to make this work for IPv6, i.e. packets from exterior nodes are not forwarded unless expressly invited by an interior node, or by the customer opting to configure the router manually to allow them to pass.)

Q: What about devices on the interior network operated/owned/under contract with the ISP or a 3rd party. A: Those devices initiate outbound flows through the gateway to "call home" for monitoring and control.

Q: How to manage the namespace?
A: The ISP offers very little, if any, DNS content service on behalf of customers, i.e. autogenerated ip6.arpa PTR records would be a luxury add-on.

I can think of other questions worth posing.

Q: What about mobile IPv6?
A: The home agent is either in the IPv6 gateway itself, or on the exterior network somewhere. It isn't in the interior network unless the customer is weird and deliberately puts it there for some damned reason, and she knows why she's doing it.

Q: What about multicast routing?
A: Customers can subscribe to global scope multicast, both SSM and ASM w/ embedded RP address. They may need to pay extra to originate global scope SSM (especially if ISP usually expects to change customer prefixes with any frequency). Originating ASM with embedded RP addresses may also be supported in a similar manner if the RP is integrated with the IPv6 gateway. The IPv6 gateway is an organizational scope multicast boundary by default.

Q: What about devices on the interior that shouldn't have global addresses? A: Don't let them accept router advertisements, and let them use only link-local addresses. (If you want them reachable through IPsec tunnel-mode terminated at the IPv6 gateway, then let the gateway offer a ULA prefix in addition to the globally reachable one. Let the IPsec tunnel be the way the ULA prefix is reached from the exterior network. Such devices ought to be configurable to refuse prefix advertisements outside the ULA range.)


--
james woodyatt <jhw@apple.com>
member of technical staff, communications engineering