>PPPoE is just normal PPP, except that the PPP frames, instead of
>getting a pseudo-HDLC encapsulation and being sent over serial lines
>(or moral equivalents), get a PPPoE encapsulation and are sent over
>Ethernet.
>It's exactly as p2p as the PPP running on top of it is. :-)

	aha I see.  Do we have the list of NetBSD network interfaces
	which support IPv6?  As far as I remember, ATM, Ethernet, gif (tunnel),
	ARCnet and ppp are covered.  Not sure about fddi (I don't have
	environment to test) and some other interfces.

>> Even if it is shared medium, if you have no neighbor other than
>> router on external interface,
>This is not the case in general.  Cable-modems, for example, are
>generally pretty much like an Ethernet: everyone on your piece of cable
>is your direct neighbor.  You just have a DHCP-assigned address rather
>than a fixed address.  (Some cable-modem connectivity providers do
>assign addresses statically, but some don't.)

	I don't see any problem even in in cablemodem situation, if:
	- you do not intend to assign global address onto cablemodem subnet, and
	- cablemodem runs RIPng

	CATV company
	  |
	==+===============+==  cablemodem subnet (no global prefix
	  |		  |		-> no need to reconfigure customer box)
	customer 1	customer2
	  |		  |
	==+==		==+==
	customer 1	customer 2
	home subnet	home subnet

	I would like to run some test to make sure, but this should be no
	problem.

>You say that "autoconfiguration" is prohibited for routers; what does
>this term include?  If it includes things like how DHCP is currently
>used, I expect this to be a crippling problem for IPv6 adoption,
>because of exactly this scenario.

	What is your worry?  At this moment cablemodem assumption seems to be
	"1 node per customer", which comes to the following configuration
	(if customer is behind customer NAT cloud, they look like single node)

	CATV company
	  |
	==+===============+==  cablemodem subnet (assign global prefix
	  |		  |		-> via router advertisement for v6,
	customer 1	customer2	   DHCP for v4
	PC		PC

	If they are really using DHCP to configure customer router, they're
	broken.  RFC2131 page 5 clearly says:
>   DHCP is not intended for use in configuring routers.

itojun