To:
Iljitsch van Beijnum <iljitsch@muada.com>
Cc:
Alain Durand <Alain.Durand@Sun.COM>, dnsop@cafax.se
From:
Ted Lemon <mellon@nominum.com>
Date:
Wed, 12 Nov 2003 14:22:57 -0600
In-Reply-To:
<3A5A0748-1542-11D8-A1D0-000A95CD987A@muada.com>
Sender:
owner-dnsop@cafax.se
Subject:
Re: DNS discovery
On Nov 12, 2003, at 12:58 PM, Iljitsch van Beijnum wrote: > Reading RFC 2462 will clear that confusion right up. There is an > explicit bit in router advertisements that tells hosts whether they > should use a stateful mechanism The stateful vs. stateless bit refers to address configuration in the sense of getting an IP address for the node, not getting an IP address to which to send DNS requests. When the stateful bit is set, you are supposed to use some stateful address configuration mechanism to get an IP address to use when communicating on the network. Stateful DHCPv6 is the only protocol that can currently be used for this, only because no other protocol has, as far as I am aware, been defined. Your error may be in thinking that DHCPv6 is always stateful, and perhaps in not understanding what stateful means (it's a somewhat unfortunate use of terminology, IMHO, but that's water long since under the bridge). What stateful means is that the node seeking an address to use sends a request to a server, which may have state on it about the node seeking an address, and may use that state to choose a configuration for the node. It may also generate new state and store it at that time. This is in contrast to stateless address configuration, where the client generates an address on the fly, without consulting any server, and where there is no server keeping track of what addresses are associated with what clients. This has nothing to do with what we are talking about right now, except that DHCPv6 and DHCPv6-lite share the same transport. I've heard you repeat several times that the stateless bit in the router advertisement can be used to say which mechanism to use, but that is completely wrong, because in using this bit in the way you propose, we would be making two changes to the behavior of the client, not one - it could, as you say, use DHCPv6 to get its DNS address, but it would also be required to use DHCPv6 to get its IP address. We are specifically trying to avoid placing any such requirement on a network - stateless addrconf has to be orthogonal to how the DNS server address is obtained, because the two things are completely unrelated. So we can not use the stateless bit in the router advertisement in the way you describe, and we do need to advance a new protocol specification in order to have a mechanism for choosing RA vs. DHCP-lite, and we don't have any pressing need to do that. So we shouldn't. #---------------------------------------------------------------------- # To unsubscribe, send a message to <dnsop-request@cafax.se>.