Paul Jakma wrote:
The proxy wouldn't pick addresses, it would only map prefixes. So the end-host inside the shimmed area could use a CGA address (thanks btw for helping me to prod my brain into grokking them offline ;) ). Ie:host uses: <prefix>:<CGA interface identifier>
But how is the host told to use that interface ID? Manual config? DHCPv6?If the host uses stateless autoconfig or temporary ("privacy") addresses they sure will not have that interface ID.
Why wouldn't it be exactly? Publishing something in DNS is no guarantee of reachability. It would be /intended/ to be reachable, via shim6.The taboo was on non-globally-unique addresses in DNS, i thought?
There has been a fair amount of discussion in DNSOP WG what to recommend or not in this space, but no consensus AFAIK.
If you publish an unreachable some applications will take a long time to connect (they might end up trying the unreachable first, and when TCP fails to connect fallback to the next in the list.) Other applications might only try the first address returned by gethostbyname/getaddrinfo.
Such applications are not much good in terms of failover for multihomed hosts, but they are the current legacy, and it might be a bad idea to make them work worse then they do today.
Even if folks agree on that, the fact that the internal hosts would see both the local/internal address, and the shimmed addresses, would mean that they might seeming randomly end up using the shimmed addresses, and such internal communication ends up going via the proxy.Not sure why that'd be so. It's an interior routing issue.
I don't understand. The hosts select which source/dest combinations to try first. The recommendation for how to do this in RFC 3484. Once the host has decided, the routing system gets to forward that TCP SYN (or whatever it is) to the destination IP address.
Erik