On Jul 31, 2008, at 5:33 AM, David Woodgate wrote:
Tom, I appreciated your feedback - thank you for some well-reasoned points.Your response highlighted for me that while we as an industry are giving some attention as to how to get networks *onto* IPv6, I'm not sure that much thought has been given as to how we will ultimately get them *off* IPv4. What will be the criteria identifying when a network can be connected to the Internet only using IPv6, without any IPv4, and be confident of at least > 95% connectivity? In what year will this point be reached?
Hi David,That's a good question, one that merits more thought by the broader community. That said, it seems like a more fruitful way to approach the question would be to identify a set of tools which which one might track the evolving share of IPv4 hosts or ASNs (or more ambitiously, "important" IPv4 hosts/ASNs) that are reachable by some "meaningful" sample of IPv6 hosts over time. Something like the Routeviews Archive with additional functions maybe...
Given the likelihood that Internet access providers will wish to adapt their access services to the evolving connectivity landscape, it wouldn't be surprising if many chose to set up the equivalent of "wardialers" to regularly evaluate reachability and performance conditions between their own IPv6 end sites and a sampling of popular online destinations. If an archive of data like this generated by commercial operators (or barring that, perhaps a similar collection of results generated by multiple RIRs and/or other neutral, noncommercial entities) could be collected in some public repository, it should be possible to compare reachability from IPv6 hosts to reachability from IPv4 hosts, and thus to develop a kind of "substitutability metric" to illuminate how close IPv6 is to being "as good as" IPv4 as a means of attachment to the Internet over time.
One danger that I see in reserving a lot of IPv4 addresses to enable IPv6 network connectivity without a clear exhaustion or sunset date is that it may in fact remove industry incentive to move off IPv4 - and I don't think we should be encouraging perpetual dual-stacking in the Internet.
I agree, but the only alternative that I can imagine is a gradual but absolutely certain globally coordinated phase-out of IPv4 entirely. I can imagine ways to engineer incentives in order to promote an outcome like this, and I can even imagine reasons why such an approach might be attractive, even to most incumbent IPv4 operators, over the long term -- but my imagination fails when trying to come up with incentives that might be sufficient to overcome near-term discomfort that such an approach would inevitably create, especially among the largest IPv4 holders.
[For example, we seem to be assuming that the responsibility IPv6- >IPv4 translation lies with a new edge or stub network that is otherwise IPv6. However, if IPv4 is ultimately to be removed, it seems to me that there needs to come a point (when the deployment of IPv6 is substantial) where it becomes the implied responsibility of the legacy IPv4 networks (or their transit providers) to do IPv4- >IPv6 translation. Otherwise, the IPv4 networks would be able to sit there for ever and just say "well, *we* don't need to change - we've been here for all these years and everyone has to connect to us as we are!". I admit that I don't immediately perceive what possible changes in market conditions would enable such a transition from the first state to the second.]
It seems that you have arrived at the same failure of imagination that I described above ;-) I would agree that prop-062-v001 "envisions" (i.e., anticipates, but might also contribute to, marginally) a long transition period. However, given the perverse incentives that will accompany any/all of the most likely / most anticipated outcomes (e.g., creation of an IPv4 transfer market, or just continuation of the status quo until the end, etc.), even a long transition is an improvement over the real possibilities of a closed industry (aka inevitable external intervention) and/or no transition, ever.
All that said, I have indicated in my previous emails that I would be more sympathetic towards proposals that explicitly linked IPv4 allocations to IPv6 planning (especially one that effectively supported multiple IPv6 hosts per IPv4 address). I do not believe prop-062 does this - it implies an intent towards that, but it does not actually specify any direct association between an applicant's IPv6 status and an IPv4 allocation, and I suggest that there is a lack of a critical assessment as to why a uniform allocation of a / 22 is the best approach.The type of issues I believe should be considered by an IPv6-focused proposal would include: - A specification of the IPv6 and general criteria on which the IPv4 allocation would be dependent. - A discussion of the criteria governing the appropriate size of an IPv4 allocation - A discussion of the amount of space that should be reserved for this proposal, with the reasoning behind that amount (assumptions, forecasts, etc.).
I'll defer to the proposal's authors on these points, but IMO they all seem reasonable and modest enough to work through before the Christ Church meeting...
- A sunset clause - a date after which any remaining addresses would be made available for allocation by justified request. This could be a far-away date like 31 Dec 2020 or longer - if we get it right, then no one will want IPv4 addresses after the nominated date, and if we get it wrong there should be plenty of time to change the policy.
To me at least, 2020 seems wildly, unimaginably optimistic based on current facts. Since the goal of the proposal is itself self-limiting -- i.e., keep some IPv4 available to new entrants for as long as IPv4 remains demonstrably non-substitutable, and not one day more, I'd prefer to see some actual empirical basis for sunsetting, e.g., something like the metric you suggested in passing above.
Additionally, it makes sense (to me at least) to leave the question of what to do with any IPv4 remaining after prop-062-v001 sunsets to be determined by community members at, or at least closer to that time. Defining appropriate sunset conditions for the policy within the policy itself seems like a good idea, but that would seem to preclude the idea of incorporating additional details that dictate conditions afterward. As you suggest, best to leave that for another day and another, separate policy proposal.
(It's just as well that the proposal deadline for APNIC 26 is past, or I'd be in danger of having to write up such a proposal! :-) )One idea that has occurred to me is that maybe APNIC should not allow members to directly request addresses from the reserved IPv4 space associated with such a revised proposal, but instead should only make such IPv4 addresses optionally available as part of an IPv6 address request. (In other words, there would be no further IPv4-only allocations from the time of activation of the proposal, only IPv6-allocations with some IPv4 addresses.) I admit I haven't had time to think through whether this is a good or bad idea.
That's consistent with the point I made in our last exchange, i.e., "To be perfectly honest, the reservation fulfills this purpose most clearly for new entrants -- i.e., "initial" allocation seekers -- but the decision to make it non-exclusively available on a one-time-only basis to all comers makes sense as a lightweight, pragmatic solution to potential complications like determining eligibility, etc."
Clearly, the proposal's authors have a slightly different/more expansive rationale in mind, so I'll look forward to hearing from them on this point. But in general I agree that the compelling justification for the policy/reservation is to facilitate IPv6 incorporation/transition, so making this connection more explicit in the text of the proposal would represent (just) a modest but useful clarification.