[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: draft-wbeebee-ipv6-cpe-router-04 comments
On Fri, 27 Mar 2009, james woodyatt wrote:
Some vendors of consumer gear seem to feel the need to make the process
of connecting their various appliances together to comprise a
functioning network into a university dissertation defense. I think the
more successful ones take a more subtle approach: they try very hard to
make it as difficult as possible to connect devices together into a
network that doesn't work. When the links between the nodes are all
CAT-5 cables with RJ-45 plugs on each end, that problem is non-trivial
and shouldn't be hand-waved away as not worthy of attention.
The solutions I have seen (and I think makes most sense) is to have a WAN
port on the home device. This is the way it works with most NAT routers
today, they usually have a single WAN port and then for instance, 4 LAN
ports.
I know some providers who offer 5 IPv4 addresses per customer and who
support customer L2 switches being connected, with the downside that all
traffic between these devices usually end up going via the ISP (because
local-proxy-arp is used for security reasons). For this reason, most
people hook up a NAT device anyway, to keep the speed up within their
home.
So, even though I agree that there should be no artificial limitations
imposed, the recommendation for the most scalable solution would be to
only support single IPv6 address on the WAN side of the customer device,
and require it to do routing and support DHCPv6-PD. This saves TCAM
resources on the ISP device which otherwise have to keep adjacancy
information for potentially a large number of customer devices.
My scenario above is what I envision for ETTH (also VDSL2 and other
ethernet framed alternatives).
--
Mikael Abrahamsson email: swmike@swm.pp.se