10,000 foot view of DNS/Sitefinder/Verisign
Howard C. Berkowitz
hcb at gettcomm.com
Wed Oct 8 13:50:57 UTC 2003
After attending the afternoon ICANN Security & Stability Committee
meeting, I realized that the issues involved fall into several
related but independent dimensions. Shy person that I am *Cough*, I
have opinions in all, but I think it's worthwhile simply to be able
to explain the Big Picture to media and other folk that aren't
immersed in our field.
In these notes, I'm trying to maintain neutrality about the issues. I
do have strong opinions about most, but I'll post those separately,
often dealing with one issue at a time. For those of you new to the
media, it's often best to put things into small, related chunks.
1. Governance issues
--------------------
Did Verisign have the right, regardless of technical merit, to do
what it did without prior warning? I'm simply saying "did they do
anything contractually or otherwise legally forbidden", not "was it
strongly counter to the assumptions of the Internet" or "were they
mean and nasty."
The news/political interest here is whether any other group should or
could have affected this, or if we need new governance mechanisms.
Has this revealed any conflict of interest issues? To what extent
should a registry be able to act unilaterally? These points are
meant to be examined here in the context of law, regulation and
governance, as opposed to the less formal points in #2.
2. Process (slightly different than governance) issues.
------------------------------------------------------
Moving away from the letter of their contracts, what should they
have done (if anything) about open comment and forming consensus?
This is vaguely making me wonder if they had evidence of
WMDs....oops, wrong controversy.
Assume they had no requirement for prior discussion. What, if any,
requirements did they have for testing and validating their approach,
given that a top-level registry is in a unique connectivity position
with special privileges.
3. Internet architectural impact (slightly different than effects on
innovation and/or effect on existing software).
--------------------------------------------------
I think it's reasonable to state that Sitefinder, and changes of
"internal" behavior, violates at least the traditional end-to-end
and robustness principles. This should be considered in the spirit
of the core vs end state discussion in RFC 3439, and the
architectural work going into midboxes.
A general question here is to what extent is it important that the
Internet be consistent with its relatively informal architectural
assumptions? Even among the newer technical folk, when teaching, I
rarely hear anyone aware of the architecture work---they think "7
layers" is the ultimate answer [1].
[1] I spent over five years of my life in OSI research, development and
promotion. We may have had the answer, but, unfortunately, we never
could articulate the question. That is a lesson here
4. Is the Internet the Web? Are all Internet users people?
----------------------------------------------------------
I don't think it's unfair to say Sitefinder is web-centric. The
current responses may be useful for people who can interact with it.
Apparently, there are patches that will help with mail response and
even anti-spamming tools.
But what of other protocols, especially those intended to run without
human intervention? What about failover schemes that employ DNS
non-resolution as an indication that it's time to pick an alternate
destination?
Is the apparent trend to move from "everything over IP" to
"everything over HTTP" a good one? _could_ it be a good one in
well-defined subsets of the Internet?
5. Effects on innovation
------------------------
Innovation and stifling innovation has come up quite a bit. If one
looks at the End-to-End Assumption, the historic perspective is that
the "killer apps" appear at the edges and depend on a consistent
center (e.g., web and VoIP, the latter with a QoS-consistent center
[2]). Development in the core tends to be more evolutionary and
subject to discussion (e.g., CIDR). Other development in the core
tends to be with the implementations (e.g., faster routers and lines).
[2] Remember that the access links to an ISP usually aren't the QoS
problem. Once you get to the POP, voice and other delay-critical
services can go onto VPNs or other QoS-engineered alternatives to
the public Internet.
Verisign says Sitefinder is innovative, and let's assume that it is.
But, if so, it's an innovation in the core, which is not the
"time-proven way". When I speak of time-proven, I certainly don't
mean that there isn't innovation -- this message did NOT reach you
over a 56 Kbps line between IMPs.
Internet Explorer, for example, has a means of dealing with domain
typos, but it is contrary to the way Sitefinder does things. IE also
does it at the edge. How do we deal with potential commercial wars
between the edge and core as far as competition for innovation?
6. Stability
------------
Assume that Sitefinder and the associated mechanisms are ideal. In
such a case, users would expect it. Unless a large number of users
learn to spel and tipe gud, these instances will be points of heavy
traffic.
What are the availability requirements to make the service
dependable? This includes clustered servers at individual nodes, as
well as distributed nodes. There has to be sufficient bandwidth to
reach the nodes, and even if the node has adequate connectivity
bandwidth, there are subtle congestion issues. It was pointed out
that wireless implementers, used to expecting a small error message
in their bandwidth-limited edge environments, are less than thrilled
about getting a 17K HTML response.
Remember, if these concepts prove themselves in .com and .net, users
will expect them in all TLDs -- or we get to the generally
undesirable situation of different behavior in different domains.
Let's assume Verisign has an adequate track record of running
reliable servers -- but what would be the requirements for a new
operator of .com and .net for people expecting the Sitefinder
functionality. In a new TLD, what has to be the support on Day 1?
A very different question is whether business models associated with
this service are sufficiently robust to be sure it stays present once
users expect it.
More information about the NANOG
mailing list