Hiroki Ishibashi <[EMAIL PROTECTED]> wrote:
I am in favor of this document for site-local usages. This document appropriately limits the use of site-local addresses, and still leaves the room for future usage of them (which we don't know).
This comment raises a basic question regarding what system design principles should be applied to the specification of IPv6.
Some people would like to specify site-local addressing in IPv6, even though we have no specific requirement for it today, because it might be useful in the future. These arguments often take the form of "Site-local addresses may enable <cool thing> in the future." Others seem to argue that we should include site-local addressing for some sort of "completeness". This argument takes the form of "It makes sense to have a unicast address scope somewhere between link- local and global, because that maps to how networks are constructed." But, in my opinion, neither of these arguments makes sense from a system design perspective. IPv4 has become ubiquitous partially because it is a simple and light-weight as possible. It is the slim center of the hourglass, the one small piece of software that all nodes have to implement to communicate on an IP network. All of the complicated, optional parts are included at other layers. Throughout the history of IPv6, we have wrestled with "second system" syndrome. We've added a lot of weight to the IP protocol, sometimes adding things that are only useful in certain situations, or for some nodes. And, in my opinion, the worst possible thing that we can do in this area is add a feature that complicates every IPv6 implementation and requires complexity at every layer of the protocol stack, because that feature _may_ have some benefits later... Site-local addressing is an interesting idea, and I think that it was worth exploring. But, at this point, we've been exploring it for several years, and we've found many problems and complexities that it causes (outlined in my document), and we haven't come up with a _single_ benefit of site-local addressing that wouldn't be better handled by a simpler mechanism. [If you think I'm wrong, please read my site-local impact document, and tell me what I'm missing.] There is a direct cost vs. benefit trade-off here, and including site-local addressing in IPv6 just doesn't make sense. I am also becoming increasingly certain that the concept of communication "scope" (for both unicast and multicast, actually) is really a routing concept, not an addressing concept, and that it is NOT best handled by the use of special-purpose "scoped" addresses. Instead, it would have been better to use only globally-unique, globally-routable addresses, and to build communication "scope" into the routing and access control policy of the network. Margaret -------------------------------------------------------------------- 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] --------------------------------------------------------------------
