[172421] in North American Network Operators' Group
Re: Ars Technica on IPv4 exhaustion
daemon@ATHENA.MIT.EDU (Owen DeLong)
Wed Jun 18 14:20:23 2014
X-Original-To: nanog@nanog.org
From: Owen DeLong <owen@delong.com>
In-Reply-To: <7C7B3A12-C0E8-4995-B330-45F0DF00FD83@dds.nl>
Date: Wed, 18 Jun 2014 11:17:29 -0700
To: Seth Mos <seth.mos@dds.nl>
Cc: nanog@nanog.org
Errors-To: nanog-bounces@nanog.org
A thought exercise for folks that think we need more network bits or =
fewer host bits or whatever...
If you went from 64/64 to 96/32, what would you do with all those =
additional network numbers?
Would you still assign /48s to end-sites or would you move that down to =
/80?
If you'd move that to /80, then do you really expect a need for more =
than 281,474,976,710,656 end sites?
Consider this... The world population is 7.1 billion, and expected 10.1 =
billion by 2100 (UN estimates).
Let's figure each person needs an end site for their place of business, =
their two cars, their home, their vacation home, and just for good =
measure, let's double that to be ultra-conservative. That's 10 end-sites =
per person or 101 billion end sites.
281,474 billion - 101 billion =3D 281,373 billion remaining /48s.
Of course, since we're giving ISPs /32s, let's assume that each ISP =
serves only 256 customers and that we therefore have a 256x =
inefficiency.
That means we would burn up 25,856 /48 equivalents, leaving only 255,618 =
extra /48s lying around.
Owen
On Jun 18, 2014, at 10:45 , Seth Mos <seth.mos@dds.nl> wrote:
>=20
> Op 18 jun. 2014, om 11:41 heeft Martin Geddes <mail@martingeddes.com> =
het volgende geschreven:
>=20
>> "IPv6 will never become the defacto standard until the vast majority =
of
>> users have access to IPv6 connectivity."
>>=20
>> It may never become the defacto standard, period. Nearly 20 years to =
reach
>> 2% penetration is a strong hint that the costs outweigh the benefits.
>=20
> To be fair, it is only now that there is considerable leverage to =
actually use IPv6 outside of a academic scope. Our company is ready now, =
and it=92s just a commercial retailer. I know we are way ahead of the =
curve but I didn=92t find it all that hard.
>=20
> I see a lot of people crying foul, still, but IPv6 capable equipment =
is readily available now, and, it is up to you if you find it worthwhile =
to purchase. The worldwide IPv6 transit network is complete and most =
ISPs can actually deliver on IPv6 if you push them for it and don=92t =
let them ship you off with =84we can=92t do it yet=94.
>=20
> As such we=92ve had IPv6 at work since 2012, and we got to talk to =
engineers and it wasn=92t really that much of a problem. Also, the free =
BGP tunnel from HE.net really is a lifesaver in getting at least backup =
peering in place, and that worked fine for over a year.
>=20
>> IP's global addressing system is broken from the outset. See John =
Day's
>> presentation "Surviving Networking=92s Dark Ages - or How in the Hell =
Do You
>> Lose a Layer!?"
>> <http://irati.eu/wp-content/uploads/2013/01/1-LostLayer130123.pdf> =
(or,
>> indeed, lots of them at once.)
>=20
> I don=92t know, 64 bits for the networks, and 64 bits for the hosts =
seems fine, although to be fair, a 96/32 split could have worked too, =
more about networks and aggregated routes, less about hosts. It=92s also =
really good that there is a =84absolute split=94 at 64 bits to designate =
the network prefix part. That makes network identifying a lot easier. I =
suppose that is where the shorter network prefix is coming from, it=92s =
easier to remember.
>=20
>> It's really all about scopes, not layers - the TCP/IP architecture is
>> divided up the wrong way, and it will never be fixed. It's an escaped =
1970s
>> lab experiment that was able to extract the statistical multiplexing =
gain
>> faster than rivals, but on a performance and security "buy now, pay =
later"
>> basis.
>=20
> I like that IPv6 is close enough to IPv4 that I can just run with it. =
That=92s not a drawback. If you understand classless subnetting you can =
work with Ipv6.=20
>=20
>> May all your intentional semantics become operational,
>> Martin
>=20
> I didn=92t find it all that hard to become operational. Not everything =
I have at work does IPv6, but that=92s not really a requirement, is it?
>=20
> I don=92t care enough for backwards compatability with IPv4, actually, =
I=92m really glad it isn=92t so failure states are much easier to =
diagnose. I can see how IPv4.2 SP2 would have subtle issues with IPv4.3 =
SP1, but there is a hot fix for that, but not for your model. SOL.
>=20
> Not very different if I must say.
>=20
> Cheers,
> Seth
>=20
>=20
>=20
>>=20
>> On 17 June 2014 23:12, Andrew Fried <andrew.fried@gmail.com> wrote:
>>=20
>>> IPv6 will never become the defacto standard until the vast majority =
of
>>> users have access to IPv6 connectivity.
>>>=20
>>> Everything I have at the colo is dual stacked, but I can't reach my =
own
>>> systems via IPv6 because my business class Verizon Fios connection =
is
>>> IPv4 *only*. Yes, Comcast is in the process of rolling out IPv6, =
but my
>>> Comcast circuit in Washington DC is IPv4 only. And I'd suspect that
>>> everyone with Time Warner, AT&T, Cox, etc are all in the same boat.
>>>=20
>>> Whether the reason for the lack of IPv6 deployment is laziness or an
>>> intentional omission on the part of large ISPs to protect their =
income
>>> from leasing IPv4 addresses doesn't matter to the vast majority of =
the
>>> end users; they simply can't access IPv6 via IPv4 only networks,
>>> without using some kludgy, complicated tunneling protocols.
>>>=20
>>> Andy
>>>=20
>>> --
>>> Andrew Fried
>>> andrew.fried@gmail.com
>>>=20
>>> On 6/17/14, 5:48 PM, Jared Mauch wrote:
>>>>=20
>>>> On Jun 17, 2014, at 5:41 PM, Lee Howard <Lee@asgard.org> wrote:
>>>>=20
>>>>>=20
>>>>>=20
>>>>> On 6/17/14 4:20 PM, "Jay Ashworth" <jra@baylink.com> wrote:
>>>>>=20
>>>>>> Here's what the general public is hearing:
>>>>>=20
>>>>> But only while they still have IPv4 addresses:
>>>>> ~$ dig AAAA arstechnica.com +short
>>>>> ~$
>>>>>=20
>>>>>=20
>>>>>=20
>>>>>>=20
>>>>>>=20
>>>>>>=20
>>> =
http://arstechnica.com/information-technology/2014/06/with-the-americas-ru=
>>>>>> nning-out-of-ipv4-its-official-the-internet-is-full/
>>>>>=20
>>>>>=20
>>>>> Can't tech news sites *please* run dual stack while they're =
spouting
>>>>> end-of-IPv4 stories?
>>>>=20
>>>> <wishful thinking=3Don>
>>>>=20
>>>> I would love to see a few more properties do IPv6 by default, such =
as
>>> ARS, Twitter and a few others. After posting some links and being a =
log
>>> stalker last night the first 3 hits from non-bots were from users on =
IPv6
>>> enabled networks.
>>>>=20
>>>> It does ring a bit hollow that these sites haven't gotten there =
when
>>> others (Google, Facebook) have already shown you can publish AAAA =
records
>>> with no adverse public impact. Making IPv6 available by default for =
users
>>> would be an excellent step. People like AT&T who control the =
'attwifi'
>>> ssid could do NAT66 at their sites and provide similar service to =
the
>>> masses. With chains like Hilton, McDonalds, etc.. all having this
>>> available, it would push IPv6 very far almost immediately with no =
adverse
>>> impact compared to users IPv4 experience.
>>>>=20
>>>> - Jared
>>>>=20
>>>=20
