Re: graceful renumbering of CPE networks

[Sander Steffann <[e-mail]>]

Hi Mikael,

> Does the bulk leasequery really put back all needed state in the relay? So there is nothing needed to have the HGW do anything?

Yep. The DHCPv6 server remembers the relay and everything it has sent through it (that has not expired) and the relay can ask for a full dump over TCP of that data. That includes all the delegated prefixes and the link-local address of the next hop (which is the DHCPv6-PD client).

> For a different scenario, in the sense of "the customer was moved to a different BNG" so there is now no state on the new router for this customer (but the HGW doesn't know), how is DHCP reconfigure done?

It depends on the technology. If the connection is PPPoE then the CPE has to set up a new PPP session, which will generate the required state. When doing this for plain ethernet then I usually configure both BNGs as DHCPv6 relays, which means that they both see the request, both forward it to the DHCPv6 server, which then replies to both, generating the required state in both. Yes, this means that the HGW gets two DHCPv6 replies to every request, but that doesn't hurt much. They just ignore the duplicate packet.

That way both BNGs get the state, and will refresh the state using lease query after a reboot. And when you have two BNGs for redundancy they have to be able to handle each other's load in case of failure anyway, so keeping all the state is a good test for that :)

Cheers,
Sander