On 6 Apr 2005, at 14:50, Fred Baker wrote:
Define "anycast address" in this question?
A single unicast address bound to multiple interfaces.
An IPv6 anycast address is indistinguishable from an IPv6 unicast address. As such, any rule prohibiting the use of an anycast address in any location is unenforceable - I have no way to identify an anycast address to apply the rule.
Correct. However, the v6 addressing spec prohibits the use of an anycast address from being used as the source address in a datagram, or being bound to an interface on a host. These two restrictions effectively prohibit the use of anycast as a service distribution mechanism.
I would note that the rule is indeed problematic to enforce, as is reflected in the various v6 services that are currently provided in violation of these restrictions.
I didn't see the draft, but I was talking with Eric Nordmark about a notion I have that IP Mobility would be very useful in anycast services, and he told me that he had actually described it a year earlier. OK, so it's his proposal... Anycast works pretty well with DNS, and presumably with similar services that do not set up long tem associations. But with anything that routinely opens a TCP session, the use of anycast exposes the user to a problem when routing changes sufficiently to change servers.
Absolutely. Anycast is not inherently unsuitable for use in distributing services over TCP, but there are certainly dangers involved in doing so. These are discussed at some length in draft-ietf-grow-anycast-00.
[...]
I think that there is a good operational reason to allow for that model, which means that one really does need to be able to use the anycast address a a source just long enough to obtain the care-of address. I think it would be good for the document to describe the scenario and impose appropriate limits (you don't talk all day to the anycast address, you only use it to find the peer and get a better address for him), but it should allow for the scenario.
I believe that use of anycast (as a brief locator service, so that subsequent long-held TCP services can be provided using unicast addresses) is in draft-ietf-grow-anycast-00. If it's not, it has been suggested by others and is in my list of things to incorporate into -01.
Joe
-------------------------------------------------------------------- IETF IPv6 working group mailing list [email protected] Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------
