Folks,
Site Local well might be a topic best discussed on an IPv6 list, but
discussions about requiring applications to know network topology are
not. The discussion, so far, suggests that some experienced folks have a
very, very different model of Internet architecture than do other
experienced folks. We should be concerned about that.
The question is whether applications should be expected to know
destination reachability. Site Local is just one example of this larger
issue.
On reviewing the discussion, it occurs to me that we really SHOULD have
applications know details about reachability to the rest of the net.
The benefits would be enormous. At the least, the networking layer
would probably get a lot simpler, though the layers above it would get
more complex.
Of course, we will also require that TCP and UDP (and all other
transports) be as well-informed as applications. And the slippery slope
will see us all tumbling down.
Nevermind that the effective result will be multiple replications of the
routing table, at different layers of the architecture, there is the
minor matter of code size and complexity that was cited and needs to be
heeded.
Let's not take some problematic mechanisms -- or mechanisms that were
fine twenty years ago but have finally started to show their age -- and
coerce them into the architecture as if they were legal precedents. The
fact that some applications have been instructed about IP addresses is a
problem -- or perhaps more kindly could be called a failed experiment,
not a bit of guidance to future designers.
If someone wants to find a "culprit" that makes mobility, multi-homing,
and renumbering problematic -- other than the simple fact that it is
just plain difficult to do on an Internet scale -- please do not forget
that TCP uses IP addresses in its connection identifies. No application
forced that.
Application level routing is something that should pertain to the
application, not to basic networking topology. One of these days, email
gateways might actually rendez vous with something a tad less primitive
than static MX records.
It's a fun topic. But it has nothing to do with IP-level addressing.
d/
--
Dave Crocker <mailto:dcrocker(_at_)brandenburg(_dot_)com>
Brandenburg InternetWorking <http://www.brandenburg.com>
Sunnyvale, CA USA <tel:+1.408.246.8253>, <fax:+1.866.358.5301>