Re: Extending a /64

[Philip Homburg <[e-mail]>]

> It's easier for the purpose of a undigressed thread of logic if
> you accept the premise. ;-) 

Indeed :-)

> It's uncommon that data-links are multi
> access anymore.  For the purpose of the argument let's assume
> routing on each port.  

That is nice to assume. Maybe it will become operational reality in 10
years.

> Ref: draft-troan-6man-p2p-ethernet-00 Even
> if you actually had a shared network, you obviously don't need a
> subnet prefix.  That's equivalent to L=0.

L=0 is not enough.

> But you agree that interface ID's cease to exist if there is no
> subnet prefix?

We have 3 interface ID concepts:
1) SLAAC
2) L=1
3) routers forwarding traffic to a subnet

No SLAAC and L=0 means that the first two are gone. To remove the third one
the router should consider the subnet is unnumbered (and will only do ND
for link local addresses). No SLAAC means that all hosts have to do 
some kind of DHCP. Unnumbered subnets mean that all hosts have to do
IA_PD or homenet.

> E.g. you could use HNCP to divvy up a /64 into DHCP pools that HNCP
> assigns to the participating routers.  Where no links has a subnet
> prefix (apart from fe80::/10 obviously). And hosts gets addresses
> assigned via DHCP.  Nodes that further wants to extend the network
> does that by participating in HNCP.

Indeed.

> The building blocks are already all there and you can extend a /64
> network, without changing or breaking the IPv6 addressing architecture.

A literal reading of the address architecture suggests that hosts are on
subnets, and subnets have 64-bit IIDs. 

To get around that, get rid of hosts. A network that consists of routers
only doesn't need subnets.

To get rid of the subnet concept, a router would have to send traffic
to the host's link local address. Conceptually that significantly changes
the semantics. It makes that relays need to maitain persistent state
(or hosts need to be part of routing protocol or hosts need to be able
to find out that soft state has been lost).