[131265] in North American Network Operators' Group
RE: Only 5x IPv4 /8 remaining at IANA
daemon@ATHENA.MIT.EDU (George Bonser)
Thu Oct 21 18:52:24 2010
Date: Thu, 21 Oct 2010 15:51:58 -0700
In-Reply-To: <AANLkTikbrPF_9gMuDKD2mMFjro4_H0cSFMHn8DN1-EVV@mail.gmail.com>
From: "George Bonser" <gbonser@seven.com>
To: "Matthew Petach" <mpetach@netflight.com>
Cc: NANOG <nanog@nanog.org>
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
> Sent: Thursday, October 21, 2010 3:08 PM
> To: George Bonser
> Cc: Ben Butler; NANOG
> Subject: Re: Only 5x IPv4 /8 remaining at IANA
>=20
> On Thu, Oct 21, 2010 at 12:53 PM, George Bonser <gbonser@seven.com>
> wrote:
> > The first step will be a registrar saying "after this date, we will
> no
> > longer issue any IPv4 addresses for whatever reason" and at the same
> > time, getting very aggressive in reclaiming space from dead =
entities,
> > hijackers, etc. =A0As time goes by, the amount of v4 space being =
routed
> > declines through natural attrition. =A0It is a combination of =
liberal
> v6
> > assignment coupled with aggressive v4 reclamation.
>=20
> Why on earth would a registrar aggressively reclaim space from
> entities if they're no longer issuing it back out?
To reduce the pool of available IPs, to reduce the reselling, transfer, =
hijacking of the space. As the amount of available v4 space declines, =
it becomes harder to obtain those resources for an operator either =
refusing to move or not wanting to move. It increases the incentive to =
move to v6 by making it increasingly difficult to operate in v4. I =
wouldn't recommend stopping the issuing of v4 space NOW, but maybe 5 =
years after runout.
=20
> Are we planning on recommending policies into the ARIN AC
> that turn ARIN into an IPv4 space reclamation entity, to hoard
> up v4 addresses?
Ok, lets say runout occurs in 2011. Set a date, say 2016 after which =
ARIN will allocate IPv6 only. The idea isn't to hoard v4 addresses, the =
idea is to stop the allocation of new blocks.
> As it now stands, the amount of v4 space being routed will trend
> towards the asymptote of maximal organizational utilization, and will
> *not* decline. Any organization that moves resources off v4 and
> frees up address space will either hold that space as an ongoing
> resource to be used for future expansions, or will sell it off on the
> transfer market for short-term cash infusions; the new holders,
> having paid good cash for it, will have a strong incentive to get it
> routed and carrying traffic as quickly as possible, to pay back
> their investment.
For a while that is true. But what will the traffic look like 5 years =
from now? If most of the major user networks are migrated to v6 by that =
time and most of the major content providers are v6, and the amount of =
native v4 traffic declines, who is going to want v4 space for anything =
new? Servicing legacy stuff makes sense but in 2016 who is going to =
roll out new deployments in v4 space? And ARIN wouldn't be preventing =
them from doing that, they just wouldn't be able to get the addresses =
from ARIN. In other words, it would be a PITA to do that and much =
easier to roll out a new deployment with v6. By continuing to allocate =
v4 space, they would be enabling the running of v4 forever.
> There is *nothing* in the system driving towards a natural attrition
> of IPv4 usage, even after runout; we simply change the allocation
> model from purely needs based, to needs+cash based.
>=20
> Unless ISPs state that they will charge additional money to
> assign v4 addresses to customers, over what they charge
> to v6 customers, there is no real pressure in the marketplace
> for the amount of v4 routing to decline. So long as the end user
> sees the same cost, and same service for using v4 as v6, there
> is no pressure towards a v6-only world.
Maybe. But look at it this way. Imagine 5 years from now a provider =
notices that only 1% of their traffic in a particular data center is v4. =
Rather than having to maintain dual-stack configurations on all the =
gear, they decide to allocate a pair of routers to v4 and go pure native =
v6 on all their customer facing stuff. Now maybe if the few people =
still using v4 want it, they can have it by tunneling 4 over 6 to that =
pair of routers. Now the vast majority of stuff in that provider's =
network is v6 only with only a couple of internal routers running v4 =
carrying the tunnels to their users who still use that space. Maybe 5 =
years after THAT in 2021 the amount of v4 traffic no longer justifies =
running v4 at all. Customers can still run v4 if they wish by tunneling =
to a v4 provider someplace else. Maybe even give the customers 5 MORE =
years to return their PA blocks, so now we are at 15 years from runout, =
the provider has reclaimed all their v4 space from their customers and =
returns it (maybe they have returned portions of that space before then) =
to ARIN and the provider no longer offers v4 services. =20
So I wasn't talking about doing such a thing immediately, I had more of =
a phased approach in mind. 5 years from runout, ARIN stops issuing IPs. =
Within 10 years of runout, providers begin to shrink their v4 support, =
possibly tunneling the traffic to a single pair of routers in their =
network, 15 years after runout, most providers can't be bothered with v4 =
support but if you absolutely have to have it, someone can get it to you =
over a tunnel from someplace. 20 years from runout most providers have =
reclaimed all their PA space and have returned it to ARIN.
> So...uh...who's going to be first to step up and tell their customers
> "look, you get a v6 /56 for free with your account, but if you want
> v4 addresses, it's going to cost an extra $50/month." ??
I wasn't talking about PA space allocated to a provider who in turn =
allocates that to customers. Providers could still issue/reclaim their =
own space as they wish. At some point, though, the tiny amount of =
traffic in that space begins to make it hard to justify having full v4 =
BGP tables at so many places. When the traffic dies to the point where =
all the provider's v4 traffic can be handled on a single pair of =
routers, it probably will be which will free up resources on the rest of =
the routers running native v6.
