> +1 > > If a better alternative is devised for 4rd, as Roland proposes here, then can > we deprecate the u/g bit usage? Seems to me that privacy addresses are > preferable anyway, if not DHCP or statically assigned ones, so any importance > assigned to u/g surely becomes a historical artifact?
indeed. 4rd tries to solve the problem of having a 4rd CE located behind the CPE (that receives the IPv6 delegated prefix). the current 4rd solution is to try to share the on-link /64 that is also used by native hosts. for a single shared IPv4 address, I believe a 4rd CE would require 2^16 addresses. (the last 16 bits are used for the CNP). i.e. a /112. the problematic assumption 4rd makes, is that it requires these addresses without having a mechanism to do duplicate address detection. what are the alternatives to a fixed reservation of the interface-id space proposed? 1) pick a separate /64 for the purpose of the 4rd function. an automated mechanism for prefix assignment is being worked on in homenet. (draft-arkko-homenet-prefix-assignment-03) 2) defend the 4rd CE address space using DAD. pick 64K addresses out ::0:a.b:c.d:CNP. chances of a conflict are small, and you would be able to detect and defend those for nodes trying to take that address after the 4rd CE comes up. I wouldn't recommend doing DAD on every address first. so you may end up in a situation where you encounter a conflict while in operation. that should be possible to handle though. 3) remove the CNP feature that makes 4rd require so many addresses to function. either of these are cheers, Ole -------------------------------------------------------------------- IETF IPv6 working group mailing list ipv6@ietf.org Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------