Re: [v6ops] CPE Prefix Sub-Delegation on LAN

[<[e-mail]>]

> -----Original Message-----
> From: Fred Baker [mailto:[e-mail]]
> Sent: Wednesday, December 22, 2010 6:12 PM
> 
> One obvious approach would be, as you say, to have the routers within a
> network enquire of an DHCPv6 server in the CPE Router requesting
> subdelegation of its PD. There is an interesting problem there in terms
> of race conditions. If I have redundancy in the network, perhaps
> something like
> 
>                                       |              |
>                                       |  +--------+  |
>                                       +--+Interior+--+
>                                       |  | Router |  |
>                                       |  +--------+  |
>                         +----------+  |              |
>       Upstream  --------|CPE Router+--+              |
>                         +----------+  |              |
>                                       |  +--------+  |
>                                       +--+Interior+--+
>                                       |  | Router |  |
>                                       |  +--------+  |
>                                       |              |
> 
> I have a problem. The upstream obviously delegates a prefix to the CPE
> Router, perhaps a /60, and the CPE uses a /64 out of that for its own
> LAN. The interior routers will then hear its RA, and will form
> addresses on the indicated LAN. They will also determine what address
> their DHCPv6 server is at, perhaps from Stateless DHCP announcements;
> this obviously wants to be in the CPE Router. They will now each ask
> that process for a prefix for their LANs. Ideally, the DHCPv6 process
> will realize that they share a common LAN and give them the same
> prefix; at least the first time that happens, I don't know how it does
> that absent configuration. I suppose it could rate limit its responses
> - send at most one DHCPv6 PD response per second, enabling one of the
> interior routers to get its configuration together and give the other
> the opportunity to discover that it needs no such delegation. I can
> also think of problems with that in a best effort network.

That's indeed an interesting problem. However, I don't see any bad consequences: end-hosts will have two working IPv6 addresses instead of one, and those IPv6 will be from different working prefixes with different working routers.

Everything else could be solved via configuration of the CPE, and I think we expect most of the users with this design to be able to log in to their CPE and configure a "cluster" or whatever it should be called. Especially when it comes to using this design for multi-homing the end hosts, then it has to be solved via configuration because no basic user will try to multi-home.

> Having sketched a problem statement in this email, let me invite a
> draft from the working group. If it actually requires new technology,
> we should send a request to 6man or somewhere else to have it developed
> (says our charter), but if it is simply operational procedure I imagine
> we can do it ourselves. I don't care who offers the draft, but let's
> get at least one on the table to discuss.

There might be some work to do to say how the dhcpv6 server in the CPE should get the prefix received from the ISP, also when the ISP is not using DHCPv6. For example, with 6rd. However, since the 6rd parameters have to be pre-provisioned or given via DHCP(v4), I can imagine that there's a way to sort this out without too much pain.

Guillaume