On Oct 16, 2014, at 8:15 AM, Lorenzo Colitti <[email protected]> wrote: > Ted, you're going in circles here. You've been arguing for many messages that > we should use ULAs because GUAs can be flash renumbered. And now you provide > an example of an event that *is* a flash renumbering, and then proceed to say > that everything will be fine because the application will reconnect.
No, what I'm saying is that in this case you can't avoid the renumbering issue, so you just have to accept what the application does. This in no way contradicts my point that flash renumbering is better avoided. > On Thu, Oct 16, 2014 at 1:28 AM, Ted Lemon <[email protected]> wrote: >> My point was that homenets should have ULAs, and should not use GUAs for >> local communication, because GUAs can be flash renumbered, >> > Actually, they can't. Yes they can, as you just agreed: > On Wed, Oct 15, 2014 at 12:13 AM, Ted Lemon <[email protected]> wrote: >> There's no reason to do flash renumbering. >> > I think you'll find that the reason is that [ hardware manufacturers support > | operators are willing to spend router resources for ] only one forwarding > entry per customer at a time. That seems to me to be begging for trouble. As a rule routers have too much memory, not too little, so I find this reasoning unconvincing. Taking a little memory away from the buffer cache to make the forwarding table bigger seems like a really good idea. > Which means that if you *want* to force it to use ULA inside the network and > GUA outside, the only scalable option is to use split-tunnel DNS. You could > change the policy table too, but most users won't, unless the standards > change, and major OSes change the policy tables. We have already talked about good reasons for doing split-horizon DNS: you don't really want to advertise all the hosts on the local wire, just the ones you want to be globally-reachable. And changing the policy table isn't that hard: if we want to have a special policy for the local ULA, we already have a mechanism for doing it that does not require O.S. vendors to hard-code a different policy table nor users to select one manually: stateless DHCPv6. And for those who consider the use of DHCPv6 déclassé, it will still work without: you just won't be protected from a provider doing flash renumbering. _______________________________________________ homenet mailing list [email protected] https://www.ietf.org/mailman/listinfo/homenet
