Re: On IETF policy for protocol registries

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



--On Wednesday, January 20, 2016 08:22 +0100 Eliot Lear
<lear@xxxxxxxxx> wrote:

> This having been said, I don't know that this is a serious
> problem with the 5785 registry, nor do I believe we should
> make changes based on a single request/response.  My
> experience is that there is no perfect policy; and actually I
> appreciate that Mark and company are willing to donate their
> time to the cause.

Agreed.  While I understand and sympathize with the various
grievances, much of this very long thread has seemed to me to be
mostly fairly minor, falling into issues that should either be
taken up with the relevant AD (as Barry suggested) or handled as
appeals if the AD is non-responsive, or as proposed amendments
to 5785, either to change the model or to conclude it wasn't
such a good idea in the first place.  In either of the latter
cases, I would have expected to see an I-D and, if needed, a
request for a separate mailing list long before this.

However, one branch of the conversation, of which the following
seems to be a member, seems to call for comment on the IETF list.

--On Wednesday, January 20, 2016 18:45 +1000 Matthew Kerwin
<matthew@xxxxxxxxxxxxx> wrote:

> On 19/01/2016 9:04 PM, "Eggert, Lars" <lars@xxxxxxxxxx> wrote:
>...
>> I'm not sure. But would you even need that? You can just make
>> foo resolve to 80.

> Because I might have strong reasons to make my service
> available under http://example.net/foo/ ... so I'd need a well
> known location that links to the relevant entry points under
> "/foo" -- be that special headers or markup in "/", or
> something in "/.well-known/", or something else.
> 
> At which point I'd have to ask, why not just use .well-known
> in the first place?
> 
> I'm assuming that either: the person minting the URL doesn't
> know my server setup (otherwise they could include the full
> path *and* the numerical port from the start, and skip the
> whole dance); or this is for resolving non-URL identifiers
> (like "foo:matty@xxxxxxxxxxx")

The above, and some postings like it, seem to me to raise two
issues that I believe we should be thinking about more broadly:

(1) The Internet has gained a good deal of robustness and,
arguably, much of its worldwide availability and utility, from
its extremely distributed nature.  If I want to do something for
myself or within my family, a small circle of friends, or even
my country, I generally don't need to get the permission of the
IETF or anyone else.  We know there are exceptions that require
some sort of central administration, at least at the top level,
with the root of the DNS and IP addresses being the examples
that are most visible to the broader community but we have
traditionally tried to minimize the number and impact of such
cases. Parts of the community have been pushing the issues of
which this is a part as "permissionless innovation".  

The importance of that robustness and distributed decision model
suggests that we need to be very careful about creating new
mechanisms that, in turn, create a greater requirement for
central administration, regardless of the style of that
administration and who performs it.  Registries for the
convenience of others in avoiding collisions but that do not
create requirements because collisions are unlikely are in a
sort of in-between area.  

In that context, these arguments about ".well-known" or anything
else that requires making global decisions about what does and
does not match the category, with probably high risk of
collisions, especially if the main criteria for wanting to be in
the registry starts with "I want to do..." suggest that the
whole idea may be, on balance, a bad one.  Given the history of
the last 10 or 15 years, it is not hard to imagine demands for
national control over names, requirements that ".well-known" be
defined for a dozen (or many more) languages other than English
(with or without restrictions to Latin script).  From my point
of view, getting into those positions are bad for the Internet
and their tendency to draw in those who main interests lie in
centralized control and policy debates are an unfortunate
symptom, but one could equally argue that the latter are the
problem and the implications for the Internet are just
collateral damage.  Either way, I do not suggest that we should
never have such things (and .well-known in URLs in particular),
only that we should be insisting on a level of justification for
them  --including analysis of whether other, less-centralized,
alternatives are available--  that is strong enough to overcome
the risks and costs.  In the case of RFC 5785, it may be time to
hold that discussion more intensely than was the case when it
was approved (rather than picking at the review and approval
mechanisms).  We might perhaps also wonder why the discussion
about principles didn't occur in more depth earlier and what
might be done to facilitate discussions of that type in the
future.

(2) It seems to me that some of the comments on this thread come
close to demonstrations of the old adage about hammers and
things that look like nails.   While it is clear that
.well-known has good and useful applications, it doesn't seem to
me to be the only way to think about some of the possibilities
that have been mentioned.    Looking at the examples above,
"foo:matty@xxxxxxxxxxx" could be a perfectly good URL in the foo
scheme.  matty.example.net could point to an DNS record with a
URI RRTYPE.  I don't remember whether urn:foo:matty@xxxxxxxxxxx
could be valid, but urn:foo:matty.example.net certainly could
be.  And it is likely that other approaches are possible, ones
that do not have the same dependencies and tradeoffs that
.well-known does (even if they have some of their own).  So I
would encourage people to try to be clear about the problems
they are trying to solve and then consider the full range of
options for solving them:  if one option seems problematic,
whether because of the technology or its administration, then
look at, or if necessary invent another one.   And, unless what
you want to invent is something like a new URI scheme bound to
some unique identity of your own, try to avoid arrangements that
require central administration, review, or approval... at least
unless you like discussions like the one we've been having.

  best,
   john







[Index of Archives]     [IETF Annoucements]     [IETF]     [IP Storage]     [Yosemite News]     [Linux SCTP]     [Linux Newbies]     [Fedora Users]