On Apr 30, 2010, at 4:21 PM, Schwarz Albrecht wrote:
> Question: If a realm is characterized by reachability, then I would
> assume that an "IPv6 realm" would equate to an "IPv6 routing domain",
> right?
well, here things get squirrelly.
Look again at the picture in my original email in this thread:
------ ------ ------ ------
// My \\ // \\ // \\ // Your \\
/ULA Realm \ / ISP1 \ / ISP2 \ /ULA Realm \
|| || || || || || || ||
| +---+ +---+ | | +---+ +---+|
| |Me | |DMZ| ------- |DMZ| |You||
| +---+ +---+ | | +---+ +---+|
|| || || || || || || ||
\ / \ / \ / \ /
\\ // \\ // \\ // \\ //
------ ------ ------ ------
|<- my realm ->|<-------- Core Realm ----->|<-your realm->|
If these were IPv4 routing domains, the counterpart to a ULA would be an RFC
1918 prefix, and the reason that it wasn't injected in the the relevant ISP's
routing was (beyond contracts and all that) that it would make routing in the
ISP ambiguous: it would be actively bad. In IPv6 routing, the reason that a ULA
is "local" is that the ULA domain doesn't advertise it outside of its domain
and the ISP doesn't accept it if it did. But if they two agreed to advertise
and believe it, there is no substantive reason that they *couldn't* - it
wouldn't hurt anything.
Now, in the default-free zone (to the extent that such a thing really exists in
the core), there are all sorts of places in which we aggregate routing,
disbelieve routes advertised, and choose to advertise or not advertise routes.
That makes it hard to argue that the fact of applying policy to routes marks
off routing domains. I tend to come back to the definition of an "autonomous
system", no doubt made for EGP or BGP but escaping search at the moment: "a set
of networks using a common routing protocol and under common administration"
for a "routing domain" - it is a region in which we can rationally expect
routing to be consistent.
_______________________________________________
nat66 mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/nat66
