[162594] in North American Network Operators' Group
Re: "It's the end of the world as we know it" -- REM
daemon@ATHENA.MIT.EDU (Geoff Huston)
Fri Apr 26 03:12:54 2013
From: Geoff Huston <gih@apnic.net>
In-Reply-To: <517A1E3C.3040102@bogus.com>
Date: Fri, 26 Apr 2013 17:12:33 +1000
To: joel jaeggli <joelja@bogus.com>
Cc: nanog@nanog.org
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
On 26/04/2013, at 4:27 PM, joel jaeggli <joelja@bogus.com> wrote:
>>=20
>> I also find it a bit strange that the runout in APNIC and RIPE was =
very different. APNIC address allocation rate accelerated at the end, =
whereas RIPE exhaustion date kept creeping forward in time instead of =
closer in time, giving me the impression that there wasn't any panic =
there.
>>=20
> apnic allocation reserved the final /8 for /22 maximal allocations. =
Couple that with some qualifying very large assignments towards the end =
of stage two e.g between feb 1 and april 14 2011 7 provider assignments =
combined soaked up more than 2 /8s and you get rapid runout towards the =
endgame.
>=20
APNIC used a 12 month allocation window right up to the point of =
exhaustion, while RIPE was operating on a 3 month window, as is ARIN. =
That may be a contributing factor in explaining the differences in =
behaviour in the final months / weeks.
But its not just that.=20
Other factors include large developing countries with massive DSL =
deployments underway (China, India) mean that in the APNIC region we =
were not looking at a wired infrastructure market sector that was =
already saturated. Quite the opposite. Similarly the wireless market in =
Asia was / is expanding rapidly for much the same reason (wireless is =
cheaper to deploy than wired if you have absolutely no pre-installed =
wireless infrastructure). i.e. the unmet demand overhang as compared to =
the available address pools was massive in Asia. Now that does not imply =
that Europe and the Middle East has no demand overhang, but perhaps not =
on the same scale as was experienced by APNIC in early 2011.
Also in September last year the European financial situation was still =
impacting on the problems of the service industry (and still is in many =
countries). So the underlying capital-driven demand factors were =
different between Europe and Asia. Perhaps it was more challenging for =
European entities to demonstrate an expansion of their Internet service =
infrastructure over rolling 3 months windows due to a slow down in =
consumer demand in parts of Europe.
What factors will play out in the North American market? It might be =
interesting to look at address allocations by country by year. One such =
table of the top 10 countries in terms of IPv4 allocations since 2007 is =
at http://www.potaroo.net/ispcol/2013-01/2012.html, table 3.The peak US =
year was 2007 with 48M addresses. in 2011 ARIN introduced the 3 month =
allocation window, and allocating that year halved from the previous =
year. Last year they were a little higher at 28M addresses. What drove =
last year's numbers in ARIN was a total of 16M addresses allocated to =
Canadian entities. So to what extent is this a saturated market already =
in terms of the deployment of service infrastructure? To what extent are =
new devices simply replacing old, and to what extent are the dynamics of =
the market in that region driven by provider churn as distinct from =
greenfields expansion? Obviously the answers to such questions have a =
strong impact on the underlying model of overall demand for more =
addresses in the region.
And of course one of the hardest factors of all: Panic is extremely =
difficult to model. Most forms of predictive modelling reach back in =
time and then use that date to push forward. but panic is of course =
different. It does not drive off past behaviour but feeds off itself. =
The APNIC runout was exceptionally hard to model at the time because the =
incidence of large allocations rose very quickly in March. Yes, I'd =
ascribe that to panic. That reaction was not so evident in RIPE in =
August / September last year. So it appears that panic, or the level of =
panic, is not a constant factor. Different regions at different times =
appear to elicit different responses to impending exhaustion.
Geoff
