RE: Comments on draft-wbeebee-ipv6-cpe-router-01.txt

["Hemant Singh (shemant)" <shemant@cisco.com>]

I and Wes are working with NTT, Japan as well and they clearly support
the WAN interface on the CPE Router to acquire a global IPv6 address.
NTT also happens to have a functional IPv6 ADSL deployment that is few
years old.  We listen more to working deployments than any requests from
folks that still not don't seem to be complete with their IPv6 thinking
for DSL.  I am meeting with NTT folks in Dublin to iron out some details
for the CPE Router - unless I close with them, our CPE Router document
is still subject to change.  We will certainly work with AT&T till we
have a plan in place for them too. 

BTW, I and Wes work in the Cable industry and build cable aggregation
routers (CMTS) for a living.  Our cable deployment has completed
standards, and CMTS and cable modems working as of 2 years back for
IPv6. Our deployment also uses a global IPV6 address for the WAN
interface of the cable modem.

I have a simple bottom-line. The CPE Router does have two separate
routing domains in the WAN and LAN interface(s).  Routing domains need
their own global IPv6 addresses and different subnets. There is also no
dearth of IPv6 global addresses, so I don't see what the problem is if
the WAN interface get a global IPv6 address.

I agree with you that Loopback interfaces are not common to home
Linksys-like routers.  But, IPv6 being a new and different space
(everything from IPv4 cannot map as is to IPv6), if a Loopback meets a
requirement for one operator, it was worth considering. 

Hemant
 
-----Original Message-----
From: Stark, Barbara [mailto:bs7652@att.com] 
Sent: Tuesday, July 15, 2008 4:39 PM
To: Hemant Singh (shemant); v6ops@ops.ietf.org
Cc: Antonio Querubin
Subject: RE: Comments on draft-wbeebee-ipv6-cpe-router-01.txt

I disagree that if the loopback interface is not configured "you will
have to use the Numbered model of our document and assign a global IPV6
to the WAN interface". Just to be sure, I think you're saying that the
CPE router will need to get a global IPv6 address via SLAAC or stateful
DHCPv6, from the access network. I do not see any reason why the CPE
router cannot use one of the addresses from its assigned prefix. Since
we are definitely considering using the "unnumbered model", and we have
no loopback interface in our devices (and have no intention of having
them), I think I'm going to have to try to get a second opinion on this.
Does anyone else have an opinion on whether or not the "unnumbered
model" will only work in the presence of a loopback interface?

By the way, I've never seen such a loopback interface in any mass market
retail CPE routers. To me, the capabilities found in such mass market
routers should be the benchmark for "very normal". I'm also not familiar
with it in DSL routers. DSL routers are also "very normal", worldwide.
I've been trying to search the Internet for products that include this
interface, and all I can find are Cisco products with IOS. I'm curious
as to whether any other vendors implement this interface, and, if so, on
what sorts of mass market products.
Barbara

-----Original Message-----
From: Hemant Singh (shemant) [mailto:shemant@cisco.com]
Sent: Tuesday, July 15, 2008 3:08 PM
To: Stark, Barbara; v6ops@ops.ietf.org
Cc: Antonio Querubin
Subject: RE: Comments on draft-wbeebee-ipv6-cpe-router-01.txt

Comments in line below between <hs> and </hs>. 

-----Original Message-----
From: Stark, Barbara [mailto:bs7652@att.com]
Sent: Tuesday, July 15, 2008 1:15 PM
To: Hemant Singh (shemant); v6ops@ops.ietf.org
Cc: Antonio Querubin
Subject: RE: Comments on draft-wbeebee-ipv6-cpe-router-01.txt

Comments in-line. 
Barbara

-----Original Message-----
From: Hemant Singh (shemant) [mailto:shemant@cisco.com]
Sent: Tuesday, July 15, 2008 11:34 AM
To: Stark, Barbara; v6ops@ops.ietf.org
Cc: Antonio Querubin
Subject: RE: Comments on draft-wbeebee-ipv6-cpe-router-01.txt

Barbara,

Since Antonio also raised a point about IPv6 addresses assigned to the
WAN interface (and the number of WAN interface(s)), we are combining a
reply to both you and Antonio. 

Barbara, when you say, 

"We have considered the possibility of a separate IP address for our
TR-069 management of the CPE routers that we supply, but would want this
to be a configurable option."

could you please elaborate. Does this separate IPv6 address get assigned
to the WAN side of the CPE Router? 
<Barbara>
One idea is that if a CPE router acquires a global address (SLAAC or
DHCPv6 on the WAN), that it would use this for the TR-069 management
interface, and it would use a global address from the prefix its given
for all other Internet communication. Some people think it would improve
security of that management interface (because of the ability in the
access network to control what traffic goes to/from that IP address).
Others consider it unnecessarily complex. The preferred approach (by
those who think having a separate management interface is too complex)
would be for CPE routers to use one of the addresses from the assigned
prefix for all Internet communication (between the CPE router and the
Internet), independent of whether it acquired a global address through
SLAAC/DHCPv6. The address acquired through SLAAC/DHCPv6 would only be
used to communicate with the access network (DHCPv6, ND, etc.).
</Barbara>

If yes, if the WAN interface is only
assigned a link-local address, then what network interface on the WAN
side does this IPv6 address get assigned to? 
<Barbara>
I'm experiencing some confusion here -- I thought that any given
"interface" could have multiple IPv6 addresses associated with it, and
it would simply need to have rules as to which address to use for which
traffic. The CPE router uses the address acquired from the access
network for DHCPv6, ND, etc., and it uses a global address from the
prefix for communication with other global addresses (not in its
prefix). Why do you believe separate interfaces may be needed?
</Barbara>

<hs>
Ah, just confusion on our side and also because we haven't read the
TR-069 document yet. We thought, it's another network interface being
discussed for management. To make sure we understand your statements
above, here is what we think you are saying:
The WAN interface gets a link-local address that is used by DHCPv6 and
ND. The WAN interface also gets a global IPv6 address (doesn't matter,
if the global is an IA_NA from DHCPv6 or an address out of the IA_PD
prefix doled out to the WAN interface) that is used for management  and
any other communication.
</hs>

One choice we have is to
assign the IPv6 address to the Loopback interface facing the WAN side.
<Barbara>
We don't have such a LAN-facing loopback interface in any of our mass
market CPE, that I'm aware of. And I don't think we intend to start.
</Barbara>

<hs>
The CPE Router is a router for which a network Loopback interface is
very normal. Other folks have asked us for this Loopback interface, so
we are going to be very careful changing the draft for the Loopback
interface. However, it's optional to be configured. But then you will
have to use the Numbered model of our document and assign a global IPV6
to the WAN interface.
</hs>

Also, when reviewers ask for another network interface on the WAN side,
or another IPv6 address on the WAN side, could reviewers please
appreciate the fact that another WAN interface can mean two things for a
router. Either we have second physical WAN interface with a new
mac-address or we still have one WAN interface and the second WAN
interface is just a logical interface bound to the physical WAN
interface such that both the physical WAN interface and the logical
interface share a mac-address between them. Antonio and Barbara, which
one of these two interfaces are you talking about? Further, if one is
spawning logical interfaces on any router, once doesn't have to stop at
one extra interface. Go ahead and spawn more if one wants. 
<Barbara>
I mostly agree with this description of logical and physical interfaces,
although it can be more complex with DSL, and I consider Ethernet PHY to
be physical, but Ethernet link layer to be logical. A single physical
ADSL connection can have multiple ATM PVCs. An ATM PVC can support
multiple PPPoE sessions (although each ATM PVC only has a single
Ethernet "interface"), in which case each PPPoE session would also be an
"interface". It's also possible to bond multiple physical DSL
connections into a single logical connection (at the ATM or Ethernet
layer). I think that in these IPv6 discussions, I'm referring to an
interface at the Ethernet link layer. I think that should be considered
"logical", and physical would be DSL PHY, Ethernet PHY, etc. I'm
definitely not referring to the PHY interface. PHY needs to be
completely independent of IPv6 (and IPv4) discussions, IMO.
</Barbara>

<hs>
Yes, the OSI model says PHY needs to be independent of IPv6 and IPv4 at
layer 3. So are we on the same page that you and Antonio need only one
Ethernet physical WAN interface, but there may, of course, be any number
of logical interfaces bound to the physical Ethernet interface. Please
let's make sure, if anyone needs more than one physical port for the WAN
side of the CPE Router - please speak up if you do.
<hs>

Further, Antonio, when you say,

"Additionally, for those vendors that wish to integrate the layer 2
(DSL/cable) modem as part of the CPE router (where the "WAN"
encapsulation is not ethernet), perhaps a separately named interface
definition might be appropriate to avoid confusion with "WAN"."

In such a case, the WAN interface is a logical interface that bridges
the CPE Router to the broadband modem. We clarified the WAN interface
definition as follows with new text

WAN interface - a single physical network interface on the standalone
CPE Router that is used to connect the router to the access network of
the Service Provider. When the CPE Router is embedded in a device that
connects to the WAN, this interface is a logical network interface that
bridges the device to the CPE Router. Some devices which can have an
embedded CPE router are: a cable or DSL modem, or a cellular telephone,
etc. 
<Barbara>
I really think that we should focus on WAN interface at the Ethernet
link layer. Each MAC address presented to the WAN is a separate
interface. Since I'd also like for this document to apply equally to
Ethernet-PHY-to-the-WAN CPE routers (including ones that may get
cascaded inside the LAN), I would recommend a more general and simpler
description, such as:
WAN interface - an Ethernet link layer interface on the CPE Router that
is used to connect the router to the access network of the Service
Provider or to other CPE routers that are between it and the access
network. 

<hs>
Good suggestion. We'll consider this text when we prepare a new
definition of the WAN interface based on other people's reply too.
</hs>

When there is a single WAN interface enabled on a CPE router, it's not
necessary to use RIP or have static routing entries to determine how to
route traffic to the WAN. 

<hs>
Not quite. What happens to the traffic from the Service Provider (SP) to
the home? The aggregation router (agg rtr) needs to know routes to the
home to send data to the home. RIPv6 is one option via which the home
sends its route information (routes to PD) to the agg rtr.
Alternatively, the agg rtr may snoop DHCPv6 message and inject routes in
the agg rtr for the PD's in DHCPv6 messages. There may be yet a third
solution that we are not going into.
</hs>

Thanks.

Hemant & Wes


When there are multiple WAN interfaces enabled, it is necessary. When
there are multiple WAN interfaces with an
IPv6 stack enabled, I would expect each of them to do SLAAC/DHCPv6 and
request a prefix. If each is assigned a prefix, I would expect each of
them to place IPv6 addresses from that prefix on the various LAN
interfaces, and to advertise that prefix or assign addresses from it to
devices on the LAN. That is, each WAN interface would meet the
requirements in this document. 

It's also possible to have multiple PPPoE sessions over a single
Ethernet link layer interface. In this case, each PPPoE session would
need to behave like a "WAN interface". But I don't see the need to go
into that level of detail in this document.
</Barbara>