My DHCPv6+PPPoE post received a very comprehensive comment from Ole Troan (thank you!) in which he explains the context in which DHCPv6 was developed (a mechanism to give a static IPv6 prefix to a customer) and its intended usage (as the prefix is static, it should have a very long lifetime).

However, when you deploy DHCPv6 in some modern access networks (it’s not just PPPoE, Carrier Ethernet fares no better), you might experience subtle problems. Let’s start with a step-by-step description of how DHCPv6 works:

• CPE router reboots. IPv6 is configured on the outside interface. We can use link-local address or SLAAC (in which case two IPv6 prefixes are consumed per customer).
• CPE router sends DHCPv6 request toward the PE-router. The DHCPv6 request includes the IA_PD option.
• PE-router receives the DHCPv6 request and either allocates the requested IPv6 prefix from a local pool (in which case the prefix is dynamic and somewhat random) or forwards the DHCPv6 request to a central DHCPv6 server (DHCPv6 relay functionality).
• In both cases, as the DHCPv6 reply goes back through the PE-router, the PE-router installs a static IPv6 route to the delegated IPv6 prefix. The next-hop is obviously the CPE router requesting the prefix.

My PE-router running IOS release 15.0(1)M did not insert the required static route when working as a DHCPv6 relay. DHCPv6 server functionality worked as expected.

So far, so good. Now imagine the PE-router reloads or its access Ethernet interface flaps. The PE-router loses all static routes to the CPE routers that were inserted in the IPv6 routing table based on DHCPv6 replies. However, the CPE routers assume everything is OK (in a typical mixed L2/L3 access network like the one shown below, a problem on one side does not result in a link loss on the other side) and try to renew the delegated prefix’s lease only when it’s about to expire. In the meantime, the customer has no IPv6 connectivity.

Fortunately, DHCPv6 implementation in Cisco IOS is pretty smart. When you use a local IPv6 pool on the PE-router, the PE-router rebuilds the static routes from the local DHCPv6 bindings. If you use a local pool and store DHCP bindings in a database, they would survive router reload as well.

It’s highly recommended to use ipv6 dhcp database to store the delegated prefixes.

But what if you decide to use a central DHCPv6 server and DHCPv6 relaying on the PE-router? How would that combination survive a link loss or a router reload? What am I missing?

Update 2012-01-19: DHCP Bulk Lease, available in Cisco IOS release 15.1(S) solves the state loss in DHCPv6 relays.