On 6 Apr 2005, at 16:09, Manfredi, Albert E wrote:
Any reason why the same rules that apply to multicast addresses wouldn't also work here?
Actually, yes.
Anycast is being used to deploy services which, to clients, appear identical to normal unicast services. The protocols used are varied, and widely deployed (e.g. DNS, HTTP).
Clients source datagrams with a destination address set to the service address, which in the cases we are discussing are deployed as anycast addresses (i.e. they are present on more than one interface, usually on different hosts). The clients have no way of knowing in advance whether the destination addresses they are using are deployed on just one interface (in which case we are talking about unicast) or more than one (anycast).
The destination address is anycast, to find the service. But the source address in the reply is the unicast address of the first server to respond, which presumably is the closest one at the time.
That might be a reasonable way to design a new protocol to be deployed using anycast (although I think the security implications might bear close scrutiny), but it's not a reasonable way to distribute services which use existing protocols.
To use DNS as an example, a DNS resolver sends a request to a nameserver using a particular destination address, and expects the reply to be sourced from the same address. If the reply was sourced from a different address, the client (resolver) would not accept it as an answer to the original request.
Joe
-------------------------------------------------------------------- IETF IPv6 working group mailing list ipv6@ietf.org Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------
