[arin-ppml] NAT444 rumors (was Re: Looking for an IPv6 naysayer...)
On Feb 22, 2011, at 3:40 AM, Owen DeLong wrote:
>> There seems to be a position, taken by others on these lists, that IPv6 is the only address family that matters. Interestingly, this position seems to be most pronounced from people not involved in operating production networks. But, regardless, if I were to accept this position then I might also agree that it doesn't matter whether or not draft-donley-nat444-impacts is misleading.
> I don't think anyone has said that IPv6 is the only address family
> that matters. What I think people, myself included, have been saying
> is that IPv6 is the only way forward that does not involve many of these
> problems. (See my earlier Titanic post).
I agree completely: IPv6 is the only way forward that avoids these problems. In fact, an understanding of CGN impacts should be enough motivation for operators and users to start deploying IPv6 immediately.
> As to whether or not it matters that people misinterpred draft-donly...,
> I'm not sure whether it actually does or not. There is no flavor of NAT
> that is particularly desirable. It's a matter of choosing the one that is
> least damaging to your environment where least damage may
> boil down to a choice between 5% and 3% remaining functionality.
I agree with your sentiment, that we should choose the least damaging solutions. Call it the "lesser evil" if you'd like.
However, I think your estimates (5% vs 3%) are backwards. CGN-based solutions work for the vast majority of network traffic today - it's the stuff in the margin that breaks, according to all test reports I've seen.
> I don't think anyone is saying IPv4 no longer matters. I think we are
> saying that effort spent attempting to make the deteriorating IPv4
> situation deteriorate less is both futile and better spent on making
> the IPv6 deployment situation better.
It's not an exclusive situation - we can roll out IPv6 while continuing to maintain our existing IPv4 connectivity, support new customers with IPv4 needs, etc. As I mentioned before, we have to support the bridge we're crossing (crumbling IPv4 infrastructure) until we're on the other side (fertile IPv6 farmland).
>> Of course, we can also rely on an IPv4 address market to avoid NAT in the more sensitive situations (i.e. situations with more sensitive users). But that's a different conversation.
> Only if you expect that you can rely on a supply side in such a market.
> I am unconvinced that such will be reliable, especially after about 6
> months of trading. This also presumes that more sensitive users can
> be defined in terms of what those users are willing (or able) to pay.
This is an interesting discussion, because the timeframe is central to everything I've commented above.
Considering RIR exhaustion (4-12 months) plus ISP exhaustion (TBD, but let's say anywhere from 1 month to 5+ years after RIR exhaustion), I expect some network providers to struggle with IPv4 address exhaustion before the 3rd quarter of 2011. On the other hand, other network providers will have enough resources to last for years - let's call that "excess supply".
By all realistic estimates, any network provider that hasn't deployed IPv6 support into their infrastructure will need anywhere from 3 months to 3 years or more - let's generously say around 18 months to the point where 60% - 80% of hosts have reached IPv6 connectivity. Just considering these facts, I think we can see why some ISPs might be interested in acquiring more addresses through 2012. And those with excess supply might be motivated (financially) by a marketplace to share their resources, to meet this need.
Further, let's consider that some network services (such as content / hosting) will need IPv4 connectivity longer than others, in order to reach the long-tail. For this category, I can see why some networks might be interested in acquiring more addresses through 2013 - 2016. Fortunately, on the other side of 2012 prices should decrease because supply goes up (as some people give up IPv4). Thus the market value of an address probably can be represented by a curve peaking in a couple years and then declining to zero a few years after that.
Feedback on this would be appreciated - but my current belief is that it's realistic to plan for a couple years of trading rather than "about 6 months".
(Side note: If we really wanted people to move to IPv6 before now, we should have instituted increasing prices for RIR-provided addresses. I posit that we just didn't have the collective balls to do this.)