Re: [v6ops] Thoughts about wider operational input

[Philip Homburg <[e-mail]>]

> 2. I think in the future almost
>    every site could want to be
>    multihomed, in some cases "n" links active, many other cases
>    just as a backup.
> 3. This means that renumbering is not
>    (probably) a valid choice in
>    any cases. 
> 4. Can we make PI work in such "huge scale"
>    scenario?
> 5. Can source-address forwarding work and solve all
>    that, or we need
>    that and/or something else.

It seems unlikely to me that PI will work in a 'huge scale', so we are
left with poor man's multi-homing where the user's network will use addresses
provided by different ISPs.

For outbound connections we have good options. Putting the smarts in the
host means that a router only has to drop it's default route to get hosts
to switch to another upstream.

With babel, etc. multi homing can be transparent to hosts.

The big advantage compared to NAT is that with NAT you would have a single
NAT box with multiple uplinks. With IPv6 you can put that in different
routers. Giving redundancy against router failure.

Obviously, companies that care about multi-homing would need to ask about
routers that support babel. 

The unsolved problem I see is inbound connections. What do you put in DNS?
Who keeps information in DNS up-to-date.