I was a little surprise this morning when I saw, following a /10 assignment to Akamai, ARIN has only 0.998 /8 equivalents remaining. This kicks off their Phase 4, which doesn’t look any different from Phase 2.
However, after weeks of slow allocations, there’s a sudden dip, which changes curves a bit. (It’s worth noting that Geoff Huston’s projections, which use a monte carlo simulation to run all possible extrapolations, have not been updated as of this writing). As always, when you think ARIN will run out depends on how much history you look at, and what curve you think fits best. That is, how much history do you think is predictive, and what curve do you think is the best prediction? In other words, outcomes change depending on your assumptions.
Look first at the longest history I chart: ARIN allocations since Y2K.
This shows the inexorable run rate of IPv4. Way down at the bottom right, you can see where I’ve drawn some trendlines using Excel. We’ve already passed the date that would have been predicted by a linear extrapolation. The allocation rate slowed, first during the 2008 recession, then again at IANA runout (2011 Feb 3) when ARIN changed from 12 month to 3 month allocations. Because of this, the polynomial curves all look wonky, tending toward ever shallower, and finally negative curves (curling up). So this isn’t especiall useful.
Look now at the run rate since IANA gave out its last IPv4 blocks.
I like this chart, because it shows how irregular ARIN allocations are: every once in a while, one or two organizations come in for a huge assignment. In this picture it looks as if ARIN will run out sometime in the fourth quarter of this year, or maybe January 2015. That seems very likely, unless ARIN’s Registration Services becomes even stingier with allocations, but a prediction of “6-9 months” isn’t incredible insight.
Interestingly, though, the charts based on six months and twelve months look about the same.
Both of these show a 6th order polynomial curve giving extreme weight to the most recent allocation with ARIN running out in a month; this seems very unlikely, and does not fit the history, even given the accelerating curves seen by APNIC, RIPE-NCC, and currently LACNIC in their last few months. The next most extreme date is January 2015, in both charts, which is consistent with the previous chart.
So what does all of this mean? If you haven’t been following along for a while, it means that in 6-9 months, ARIN will no longer be able to give out IPv4 addresses. Which means ISPs, CDNs, and web hosts won’t be able to add additional customers without sharing their IPv4 addresses. One would hope that all of the companies getting large assignments recently would be ready to use IPv6 for everything. If not, they’ll be buying addresses, or sharing addresses among customers (NAT).
For those keeping score on my predictions, I said:
- – ARIN would be right around a /8 in May. Maybe they’re a week earlier than I predicted, or maybe we can agree they’re still “right around” a /8.
- – LACNIC would hit its last /9 in April. They’re at 0.66 /8, or a /9 plus a /10. It’ll be another three weeks before they hit /9.
Not bad. We’ll see how the rest of the year plays out.