[137351] in North American Network Operators' Group
Re: IPv6 mistakes, was: Re: Looking for an IPv6 naysayer...
daemon@ATHENA.MIT.EDU (Owen DeLong)
Fri Feb 11 15:00:07 2011
From: Owen DeLong <owen@delong.com>
In-Reply-To: <4D554F29.8070903@ispalliance.net>
Date: Fri, 11 Feb 2011 11:56:38 -0800
To: Scott Helms <khelms@ispalliance.net>
Cc: nanog@nanog.org
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
On Feb 11, 2011, at 7:00 AM, Scott Helms wrote:
>=20
>> I don't know about that. Yes, v4 will be around for a long time but
>> considering the oligopolies we have in both eyeball and content
>> networks, ones a dozen or so very large networks switch, there is the
>> vast majority of Internet traffic right there. It will be around for =
a
>> very long time handling a tiny bit of traffic.
>>=20
> Agreed, V4 traffic levels are likely to drop and stay at low levels =
for decades.
I don't think it will be just a drop in traffic levels. I think that it =
will not be long
before the internet is an IPv6 ocean with islands of IPv4, much like it =
was
an ocean of IPv4 with islands of IPv6 years ago.
>> Facebook alone accounts for 25% of internet traffic in the US. =
Netflix
>> is estimated to be over 20% and YouTube at 10%. So that's 55% of
>> Internet traffic right there. At the other end of the transaction =
you
>> have AT&T with 15.7 million, Comcast at 15.9 million, Verizon at 9.2
>> million and Time Warner at 8.9 million (early 2010 numbers). That's =
50
>> million of the estimated 83 million US broadband subscribers. So =
once
>> three content providers and four subscriber nets switch, that is over
>> 25% of US internet traffic on v6 (more than half the users and more =
than
>> half the content they look at).
> Comcast, nor the other large MSOs, are not as monolithic as they may =
appear from the outside. In most cases the large MSOs are divided into =
regions that are more or less autonomous and that doesn't count the =
outlier properties that haven't been brought into the fold of the region =
they are in for various, usually cost related, reasons so don't expect a =
large block of any of those guys to suddenly be at 60% of their users =
can get IPv6 addresses.
>=20
I think you'll be in for a surprise here.
> While Facebook working over IPv6 will be a big deal you won't get all =
of their traffic since a significant fraction of that traffic is from =
mobile devices which are going to take much longer than PCs to get to =
using IPv6 in large numbers. Also, Netflix is even more problematic =
since the bulk of their traffic, and the fastest growing segment as =
well, is coming from Xboxes, Tivos, other gaming consoles, and TVs with =
enough embedded brains to talk directly. Those devices will also =
seriously lag behind PCs in IPv6 support.
>=20
I think you'll be in for a surprise here, too. The 4G transition is =
already underway. For the vendors where 4G means LTE, IPv6 is the native =
protocol and IPv4 requires a certain amount of hackery to operate.
In the WiMax case (Gee, thanks, SPRINT), things are a bit murkier, but, =
I think you will see WiMax go IPv6 pretty quickly as well.
Yes, it will take a little longer to retire the 3G system(s) than many =
other parts of the internet, but, I think you will see most of it going =
away in the 5-7 year range.
>> I don't think the growth of v6 traffic is going to be gradual, I =
think
>> it will increase in steps. You will wake up one morning to find =
your
>> v6 traffic doubled and some other morning it will double again.
>=20
> They'll be jumps, but they will be fairly smallish jumps since both =
the content maker, the ISP, and the device consuming the content all =
have to be ready. Since I don't imagine we will see any pure IPv6 =
deployments any time soon many/most of the IPv6 deploys will be dual =
stack and so we are still at the mercy of the AAAA record returning =
before the A record does.
You misunderstand how getaddrinfo() works under the hood. The code =
itself first does an AAAA lookup and then does an A lookup. DNS does not =
return both record sets at once. If there is an AAAA record, it will =
return first.
Some OS have modified things to resort the getaddrinfo() returns based =
on the perceived type of IPv6 and IPv4 connectivity available as an =
attempt to reduce certain forms of brokenness. However, even in those =
cases, you should get the AAAA first if you have real IPv6 connectivity.
Owen
>=20
