Ofer Inbar wrote:
I think today's implied identity of "private address space"
and "locally scoped addresses" is really muddying the
discussion. If we have a world where anyone who asks can get
a unique address block assigned to them to use for their
private network... THEN what are the remaining reasons why
people would want to use site-local addresses?
offered with the intent to grow into an answer to the last part.
Second, the IPv6 SL mechanism exposed the fundamental architectural
discrepancy between the traditional application world viewpoint that an
address was a valid identifier, and the deployed network where addresses
are topology locators and have different visibility in different regions
(aka: scopes). Almost all of the complaints against SL will persist even
with a non-ambiguous address space (the exception being connecting to
the wrong node, which was arguably unlikely to begin with given mac
based IIDs). At their core, these complaints are about bringing apps
(actually every process that resolves a data structure into a topology
locator) into the world of reality where they need to recognize that
various members in a list of potential addresses for a node may not be
equally accessible from every other node in the network. The fact that
IPv6 nodes can simultaneously have addresses of mixed scopes makes the
problem more acute, but it is no different than a multi-homed IPv4 node
where one or more of its addresses are not visible in the same scope of
the routing system, or there are different access controls placed on the
set by the network manager. The system as a whole needs to be revisited
in the context of making sure every node can resolve a given data
structure into a topologically appropriate locator.
The IPv6 WG discussion about a specific prefix is probably best dealt
with there, but the broader issue about assumptions of a flat routed
space vs. the reality of limited topology scopes is something the IETF
as a whole needs to deal with, because it exposes many historically