Define "Runout"

This week, ARIN allocated another /12, bringing their inventory to less than half of a /8. Since they allocate based on 3-month need, this is "Last Call" for allocations. My friend @toreanderson pointed out that runout date varies based on the size block you need--something I've noted before, but haven't detailed.

So I built a spreadsheet. What if allocations for the past six months are repeated? For each assignment, I decremented ARIN's inventory for that size block. If there were no blocks of the appropriate size, I decremented the next larger available block, and incremented the size allocated. In other words, if there was an allocation of a /20, but zero /20s available, I took away a /19, assigning one /20 and leaving a /20 available. If there was no /19, I would take away one /18, leaving one /19 and one /20.

The important dates:

  • 2015-01-17 Last /10 broken up
  • 2015-02-13 Last /11 broken up
  • 2015-03-09 Last /12 broken up
  • 2015-03-26 Last /13 broken up
  • 2015-04-01 Unable to fill request for /13. Allocate /14 instead
  • 2015-04-03 Unable to fill request for /18. Allocate /22 instead
  • 2015-04-06 Unable to fill request for /22.
ARIN's Minimum allocation is /24, and Minimum assignment is also /24. As of 2015-04-06, there will still be about 1200 /24s in inventory. If organizations keep applying for address space at the same rate, and getting just a /23 or /24, it may take months for ARIN to hit zero addresses. It is much more likely that organizations that have been getting larger blocks will get ARIN's approval, then go to the IPv4 address market to meet their needs.

The exact dates for each event will be different, of course. But if organizations keep applying at the same rate they have been, the middle-sized blocks (/21-/14) will be gradually drawn down. Once the last /13 goes, the medium blocks will run out very quickly. The early April date is very close to both my own current projections (regardless of history or projection curve) and Geoff Huston's.

Everyone likes visualizations, so here's a stack chart of remaining inventory, depleted as described above. Each color shows the number of remaining addresses remaining in that prefix size. We can see that the /10 is more addresses than all the rest, until it's gone. Note that each bar is an allocation, not a date, so the X-axis does not proceed linearly.

As always, I'm very interested in others' interpretations of this data. Follow and Tweet me!


Lee Howard
13 Dec 2014
@WLeeCoyote