> Of course you can do that. However, then there will be some applications > that try to access anycast addresses (perhaps on the same link) that will > instead bother some other computer that may even run the same application > as an anycast host might do.
For the currently defined use of anycast (the subnet anycast which isn't used by anything and the home agent anycast) this is not a problem. Even if a HA is attached to such a link you can avoid the problem by not configuring the MNs to have that as their home link. > > Working the same as IPv4 practice can't be a disadvatage. > > I'm sure you meant that in the context of this very specific discussion, > not as a general design rule. > > Otherwise we'll have to start building IPv6 NAT boxes right away. Stated with a bit more detail to clarify what I mean it does make sense as a general design rule. If there is IPv4 designs, protocols, and/or operational practises IPv6 should follow those unless there is a good reason. Thus IPv6 shouldn't choose to do things differently just because it can be done differently but instead look very carefully at the cost of being different than IPv4 (in terms of mindshare, training, etc) and the benefits. > Still no one has pointed out any downside on reserving the space by > restricting prefix lengths. And I don't think anybody has pointed out the benefits and argued that those benefits exceed the cost of training folks that e.g. a /31 in IPv4 corresponding not to a /127 but to something else. > Robert Elz's point was the best/only one so > far, which is that fewer rules are better. But the rest of your note, > which I didn't quote, amounts to some more rules. And I don't see > why applications should have to know rules about subnet prefix > lengths in order to work better: > How many "applications" do we expect to use subnet reserved anycast addresses? (and what is your definition of "application"?) Mobile IPv6 uses them but that is IP infrastructure and not an application in my notion of the meaning of "application". > Plus, why shouldn't a host on a point-to-point link be able to run some > nifty applications that anycast is supposed to enable? If you say that > there aren't any yet, then that's a strange reason given that IPv6 doesn't > exactly have killer apps yet. We are not talking about arbitrary pt-pt links. We are talking about pt-pt links between routers that have been explicitly configured to save on address space. Thus if folks want to run those applications don't configure the pt-pt link that way. Erik -------------------------------------------------------------------- IETF IPng Working Group Mailing List IPng Home Page: http://playground.sun.com/ipng FTP archive: ftp://playground.sun.com/pub/ipng Direct all administrative requests to [EMAIL PROTECTED] --------------------------------------------------------------------
