[6bone] pTLA for Teredo testing - review closes 7 August 2002

[Thomas Narten]

Christian,

> Both, kind of. I am ready to sign off for something that says "Teredo
> servers will be turned of at some predictable date". We could debate the
> date; if I had to choose one, I would say 12/31/2006.

That's a rather long-running experiment, IMO. Hard to see how this is
different than just "deployment".

> There are two ways to stop the experiment once it starts: stop the
> servers, and junk the prefix so nobody routes these packets.

And if clients are still using it, they will just be left hanging?
What assurances are there that there won't be significant numbers of
clients still wanting/trying to use the prefix?

Let me step back a bit and summarize my general concern.
draft-ietf-ngtrans-shipworm-05.txt was forwarded to the IESG, and
there was quite a lot of strong concern fed back about particulars of
the technology. It is unclear what has changed in the document to
address that feedback or whether those who raised the concerns agree
that their concerns have been adequately dealt with. (I am aware of no
followup discussion on the revised draft.)

One issue I spoke about (and it was just one of the issues, and one I
chose to expand on in email at the time) concerned use of anycast
addresses for source addresses. My understanding from an earlier note
is that this sort of usage has been reduced in the latest draft, but
has not been eliminated. So I can't say that that specific concern has
been adequately addressed. (I haven't re-reviewed the document to
fully understand what the current restrictions are.) 

As a general statement, I don't believe the community should be
blessing open-ended experiments in which significant concerns about
the technology behind the proposed experiment/usage have been
raised. If the experiment were designed to better understand some
aspects that can only be determined through actual testing (while
folks were generally in agreement that the testing itself poses no
dangers), that would be one thing. But I don't sense that this is the
case here. In particular:

 - there is no clear description of what the experiment is intended to
   determine, how it will be terminated in a reasonable time frame,
   what the criteria is for deciding whether the experiment is a
   success or failure, how the scope of the experiment will be limited
   to some known size, and how we can ensure that the experiment
   doesn't just become actual deployment even if the experiment
   produces negative results.

 - more importantly, there are oustanding technical concerns with the
   shipworm/teredo work. I assume an attempt at addressing those
   concerns has been made in the revised draft. But until the draft
   has received adequate review and there is consensus that the work
   should go forward, I do not believe we (either IANA and 6bone)
   should be giving out a prefix for testing. To do so would
   short-circuit the community review process. I might add, a cynic
   might read the request for a 6bone allocation as an attempted end
   run around the fact the IESG has not approved the document, and
   hence, IANA won't yet allocate a prefix.

General comment. Don't take the above as a statement that I believe
that the problem is unimportant. I believe it is, as a solution to the
general problem could significantly help IPv6 deployment. At the same
time, however, a flawed solution could well be a disaster
in-the-making for IPv6 and set back deployment efforts overall. It is
precisely this concern that has prompted the pushback on the document
so far from myself and others. Thus, I believe it would be most
constructive to focus on the issue that have been raised and what can
be done about them.

Folks should also look at the IAB's
draft-iab-unsaf-considerations-02.txt. It provides some very relevant
discussion of some of the issues.

Thomas