Designing IPv6 network guidelines?

[itojun@iijlab.net]

>> 	to stabilize, try to remove tunnels and move to IPv6 over leased
>> 	line (or IPv4/v6 dual stack connectivity over leased line).  it works
>> 	for me.
>Tunnels between client boxes (the first hop) has some problems, but I
>can't see too many with tunneling over IPv4 between the sites, rather than
>having to build these with ATM PVC's or the like (for the ease, and not
>having to stick to ATM technology).
>Or are there some problems with this approach?

	from my experience: if the tunnel (IPv4) path goes over multiple
	administrative domains (like multiple ASes) I see unstabilized tunnel
	connectivity due to IPv4 unstability.  In my opinion, tunnels has to
	be very very short (I mean, less IPv4 hops), or has to be replaced
	by native IPv6 connectivity (with additional leased line or whatever).

>> 	you don't even need site-locals for point to point links.  they just
>> 	work fine with link-local address.  all routing protocols should run
>> 	fine with p2p with link-local address only.
>Routing protocols, yes.  But if you want to bind e.g. EBGP to the
>interface address rather than loopback, this might be a problem.

	I don't really agree with the above, I have not experienced problems
	with my approach.

	There are OS/router implementations that cannot establish BGP TCP
	connection (port 179) over link-local addresses, but it is
	implementation problem in specific implementations.

>Also, why _would_ traceroute work?  If the link local address of a P-t-P
>link were to use private addresses, you couldn't trace through the
>internet.

	if intermediate routers employ weak host model, traceroute will work
	just fine even if you use link-local address to p2p interfaces.
	if intermediate routers employ strong host model, traceroute will
	work only if they have global IPv6 address onto every interfaces.
	i think you have strong host model in your mindset...

itojun