Iljitsch van Beijnum wrote: > > On 24 okt 2003, at 18:14, Hans Kruse wrote: > > > 2. Several folks stumbled over the wording (in section 1.0) that > > "applications may treat these address[sic] like global scoped > > addresses". How about: > > > "Applications may treat these addresses like global scoped addresses; > > such applications will function correctly within the reach of the > > local addresses. Sites using a mixture of Globally Routable and Local > > addresses may experience sub-optimal application behaviour, see > > sections 8.0-10.0 for further discussion". > > I think this will only confuse people who aren't aware of all the > details. But the problem is more fundamental than wording issues > anyway. The problem is that there are two types of uses for local > addresses: > > 1. To number systems/interfaces that are only accessible from withing > the local network. > > 2. To have stable addresses for systems/interfaces regardless of > intermittant external connectivity and renumbering.
It's true that we *know* these addresses will be good for 1., and we don't know whether they will be useful for all aspects of 2. So the draft could indicate this. > > In the case of 1. the scope is site local, although the difinition of > "site" may be subject to change. Being able to route these addresses > throughout the internet would be more of a drawback than a usable > feature, as packets using these addresses may not enter or leave the > network. In the case of 2. having the addresses be globally routable > throughout the internet would be extremely desirable: having addresses > that are stable within the site is good, having addresses that are > stable throughout the net is even better. > > If we recognize that unique local addresses will be used in both > capacities and there is a significant chance that a locator/identifier > separation mechanism could make these addresses globally usable (if not > immediately routable), then it's obvious there are going to be > problems. For instance, coming up with relatively simple filters that > accommodate both uses of the addresses at the same time would be a big > challenge. Having to change the default filtering policies that come > with the OS for huge numbers of boxes would be another issue. Maybe, but it's what people operating Net 10 VPNs do today (assuming the boxes you mean are routers). > > I think that either it must be explicitly stated which type of > addresses we're talking about here. It would probably be best to only > specify type 1 and see what can be done for type 2 with > locator/identifier separation mechanisms. Let's exclude nothing. We don't need to be categorical about "type 2". Brian -------------------------------------------------------------------- IETF IPv6 working group mailing list [EMAIL PROTECTED] Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------