Re: site-local != NAT
2003-04-30 08:51:21
--On Wednesday, 30 April, 2003 07:09 -0700 Michael Thomas
<mat(_at_)cisco(_dot_)com> wrote:
Keith Moore writes:
> > Well, the pragmatic reality is to avoid using IP
> > addresses as substitutes for endpoint
> > identitifiers. Or maybe that's not pragmatic, but
> > tough noogies; using IP addresses as endpoint
> > identifiers is sloppy and generally a bad idea.
>
> ...and at this point the argument repeats, without your
having acquired > any additional clue.
Maybe if you and Randy stopped playing elliptical
word games we could have some communication
here. If you have a point, make it.
Mike,
I disagree that they are playing word games. But let me try a
different approach to the issue, partially to try to explain to
you, at a bit more length, what I think they are talking about.
One of the key issues in this discussion has been whether it is
reasonable to remove a previously-defined and standardized
feature, especially one that has been widely deployed for some
time, without strong justification. I think there is fairly
wide agreement that is a bad idea, even if reasonable people
might disagree as to which of those considerations apply to SL
and to what degree.
Now, like it or not, we have a number of widely-distributed
applications and other things, including some security
protocols, that use IP addresses as endpoint identifiers. They
do that, in part, because there is no more satisfactory
candidate for an endpoint identifier. Certainly DNS names, with
the current definition of the DNS, do not qualify as a
satisfactory alternative and the number of standards-track
alternatives that have received even moderate acceptance is,
well, very small. Comparatively speaking, the number of
network-seconds (or almost any other plausible measure you can
think of) consumed by the collection of applications and other
things that use IP addresses as endpoint identifiers
significantly exceeds the number of network-seconds used for
either referencing IPv6 addresses or using other types of
standardized endpoint identifiers... keep in mind that, from at
least one perspective on the situation, TCP is one of those
things that uses IP addresses as endpoint identifiers.
And, of course it is trivial to make up names, and maybe even to
pass them around. But there is quite a difference between that
statement and a protocol that really uses such names as endpoint
identifiers, carefully defines what those names are bound to and
how and when they are unbound or refreshed, how they are exposed
to the routing infrastructure without unacceptable layering
violations, etc. Try doing some serious I-D writing and working
through of the cases and transition issues.
So...
(i) If you want to get rid of the use of IP addresses as
endpoint identifiers, let's see an I-D that describes a
reasonable alternative, and how to get there from there,
processed along the standards track. Asserting that IP
addresses are not very good endpoint identifiers is not
especially useful -- almost everyone agrees with you, but many
of us are concerned about an operational network, and not, when
it conflicts, philosophy.
(ii) Don't bother arguing that SL should be retained because it
is standardized and entrenched and then turn around and argue
for getting rid of IP addresses as endpoint identifiers without
recognizing that usage are even more standardized and entrenched.
I continue to believe that there are serious and significant
architectural issues underlying this discussion, ones that we,
as a community, need to address and address soon. But, IMnvHO,
complaining about those nasty applications that use IP addresses
in ways that are, retrospectively, problematic is not only not a
solution to the problems, but adds a lot of noise that gets in
the way of looking at the architectural issues.
We now return you to your regularly-scheduled word games,
sniping, name-calling, hyperbole, and flights of fancy :-(
john
|
|