Pools and IPv6

[Paolo Prandini]

 > > >Am I then forced to have a single point of failure?
 > > >Or is there a way to propagate leases to a backup server other than the>
 > > >failover mechanism available for IPv4?
 > >
 > > Well I suppose the obvious answer is that you have much more address
 > > space and so it shouldn't be a problem to have 2 (or more) servers
 > > with non-overlapping ranges - each sufficient for the entire client
 > > base.

 > My thoughts too. Presumably the clients use dynamic addresses, and
 > since they renew at around half lease time, they still have plenty
 > of time left to change to a new address from the other server.

 > I don't really see the big need for failover with IPv6, except for
 > the possible DDNS issue. But then IPv6 DNS is another ball game in
 > the first place (not obvious that reverse DNS make much sense, for
 > instance).

I think the point here has been missed.
We are an ISP, we would like to hand out IPv6 addresses like we do
with IPv4 ones.
Given a certain UID, the address given to the router and the subnet given
to the customer (/48) have to stay the same even if something happens
to our DHCP server. Let's imagine a customer that has a network of some
10s or 100s PCs, all addresses have to change even those of internal
servers if we change the subnet.
This is not a problem with IPv4 because usually people has NAT and uses
private addresses; I know that with IPv6 you could use a link local
address but then if you have internal servers how can you reach them if
they change address, because the link local is simply not routed, from
an external point of view it doesn't exist and cannot be used in any way.
So: addresses released via DHCP must stay statically assigned to routers.
I know you can tell me : don't use DHCP for that, but we use this system
for autoprovisioning in IPv4, why not in IPv6?
I will have to research more on the argument, eventually we will modify
the DHCP daemon in some way.
Thanks anyway
Paolo