[182364] in North American Network Operators' Group
Re: Dual stack IPv6 for IPv4 depletion
daemon@ATHENA.MIT.EDU (Owen DeLong)
Wed Jul 15 16:06:46 2015
X-Original-To: nanog@nanog.org
From: Owen DeLong <owen@delong.com>
In-Reply-To: <CANjVB-hktATZdcHLhUgEuZx5nYkSCixvYR-xsQiKC76W9QRJZw@mail.gmail.com>
Date: Wed, 15 Jul 2015 13:06:37 -0700
To: George Metz <george.metz@gmail.com>
Cc: "nanog@nanog.org" <nanog@nanog.org>
Errors-To: nanog-bounces@nanog.org
>=20
> I wasn't intending yourself as the recipient keep in mind. However, IS
> their premise wrong? Is prudence looking at incomprehensible numbers =
and
> saying "we're so unlikely to run out that it just doesn't matter" or =
is
> prudence "Well, we have no idea what's coming, so let's be a little =
less
> wild-haired in the early periods"? The theory being it's a lot harder =
to
> take away that /48 30 years from now than it is to just assign the =
rest of
> it to go along with the /56 (or /52 or whatever) if it turns out =
they're
> needed. I personally like your idea of reserving the /48 and issuing =
the
> /56.
Yes=E2=80=A6 Their premise is wrong. We=E2=80=99re not talking about =
/48s for soda cans.
We=E2=80=99re talking about /48s for end-sites=E2=80=A6 Households, =
individual business
buildings, etc. The soda can is probably just one host on a /64 =
somewhere
within that /48. Every six-pack in your house probably fits in the same =
/64.
>=20
>=20
>> So you asked an interesting question about whether or not we NEED to =
give
>> everyone a /48. Based on the math, I think the more interesting =
question
>> is, what reason is there NOT to give everyone a /48? You want to =
future
>> proof it to 20 billion people? Ok, that's 1,600+ /48s per person. You =
want
>> to future proof it more to 25% sparse allocation? Ok, that's 400+ =
/48s per
>> person (at 20 billion people).
>>=20
>> At those levels even if you gave every person's every device a /48, =
we're
>> still not going to run out, in the first 1/8 of the available space.
>>=20
>> Split the difference, go with a /52
>>>=20
>>=20
>> That's not splitting the difference. :) A /56 is half way between a =
/48
>> and a /64. That's 256 /64s, for those keeping score at home.
>>=20
>=20
> It's splitting the difference between a /56 and a /48. I can't imagine
> short of the Nanotech Revolution that anyone really needs eight =
thousand
> separate networks, and even then... Besides, I recall someone at some =
point
> being grumpy about oddly numbered masks, and a /51 is probably going =
to
> trip that. :)
It=E2=80=99s not about the number of networks. It=E2=80=99s about the =
number of bits available to
provide flexibility in automating topology to create pluggable network =
components
that just work.
We=E2=80=99ve only begun to explore the new capabilities of DHCPv6-PD =
within a site.
Clamping down the size of a =E2=80=9Csite=E2=80=9D to less than the =
original design-criteria considered
when DHCPv6-PD was developed will, in fact, stifle innovation in this =
area most likely.
> I think folks are missing the point in part of the conservationists, =
and
> all the math in the world isn't going to change that. While the... =
let's
> call them IPv6 Libertines... are arguing that there's no =
mathematically
> foreseeable way we're going to run out of addresses even at /48s for =
the
> proverbial soda cans, the conservationists are going, "Yes, you do =
math
> wonderfully. Meantime is it REALLY causing anguish for someone to only =
get
> 256 (or 1024, or 4096) networks as opposed to 65,536 of them? If not, =
why
> not go with the smaller one? It bulletproofs us against the unforeseen =
to
> an extent.=E2=80=9D
The problem is that the conservationists are arguing against /48s for =
soda
cans while the libertines are not arguing for that.
/48 protects against the unforeseen pretty well. Even if it doesn=E2=80=99=
t, that=E2=80=99s why we
also have the /3 safeguard.
If we turn out to have gotten it wrong=E2=80=A6 If it turns out that we =
use up all of the first
/3 before we run out of useful life in the protocol, then I=E2=80=99m =
all for coming up with
different policy for the remaining /3s. Even if you want to argue that =
we have to
keep handing out addresses while we develop that policy, I=E2=80=99ll =
say you can hand
out the second /3 in the same way while we develop the more restrictive =
policy
for the remaining 74.9999999999% of the total address space.
Short of something incredibly surprising, we=E2=80=99re not going to =
exhaust that first /3
in less than 50 years even if we give many /48s to every end site on the =
planet.
In case of something really surprising, it would be even more surprising =
if we
found a way to make IPv6 scale to that on many levels other than address
space:
1. IPv6 developers punted on the routing problem and =
insisted that
aggregation was the desired alternative to solving the =
routing problem.
In reality, I think this will most likely be the first =
scaling limit we see
in IPv6, but I could be wrong.
Aggregation is somewhere between inconvenient and =
infeasible as
we have seen with IPv4 where we managed to get some =
small
gains when aggregation was first introduced and we took =
the
routing table from ~65,000 entries to ~45,000. Look =
where it is now.
More than 500,000 entries and continuing to grow.
More multihoming and more individual PI prefixes are =
going to become
the norm, not less.
2. IP is a relatively high overhead protocol. Putting an =
IPv6 stack into a
molecule-sized device is, in fact, unlikely because of =
the limits of storage
for software in a molecule-sized device. Nanobots ala =
Big Hero-6,
sure=E2=80=A6 Do you really think they needed more than =
a /48 to address
every nanobot on the screen throughout the entire movie? =
I bet you
could probably give a unique address to every nanobot in =
every frame
of the movie and still not burn through a /48.
So of the 400 /48s we can give each person on earth, =
let=E2=80=99s dedicate
100 of them to running their nanobots. We=E2=80=99ve =
still got 300 /48s available,
of which 1 will run the rest of their household and we =
have 299 in reserve.
Yes, their premise is wrong. They are optimizing for the wrong thing.
> As an aside, someone else has stated that for one reason or another =
IPv6 is
> unlikely to last more than a couple of decades, and so even if =
something
> crazy happened to deplete it, the replacement would be in place anyhow
> before it could. I would like to ask what about the last 20 years of =
IPv6
> adoption in the face of v4 exhaustion inspires someone to believe that =
just
> because it's better that people will be willing to make the change =
over?
I try to avoid =E2=80=9C20 years=E2=80=9D, but instead say =E2=80=9C50 =
years=E2=80=9D. The reality is that in any wildest
projection of current policies and crazy scientific development, the =
first /3 lasts more
than 100 years.
If you don=E2=80=99t think IPv4 is past its useful lifetime, you =
aren=E2=80=99t paying attention. However,
even assuming that IPv4 (RFC791, September, 1981) lasts another 10 =
years, that
would be a total protocol life time of (2025-1981 =3D) 44 years. Yes, =
there will still be
islands of IPv4 utilization floating around well past that time just as =
there are still
islands of NCP today. Oh, wait=E2=80=A6 Just as there are islands of IPX =
today.
However, nobody will route IPv4 natively on the global internet 10 years =
from now.
It will have gone the way of latin. Scholars will remember it, but it =
will not have much
in the way of day-to-day utilization except in specialized =
circumstances.
I don=E2=80=99t think people will make the change just because it=E2=80=99=
s better. I think just as is the
case now, somewhere within the next 50 years, people will move away from =
IPv6
because we hit some other limitation baked into the protocol and we had =
to move on.
As long as we have enough addresses to make sure that address shortage =
is not the
cause of protocol end-of-life, any additional conservatism is, in fact, =
just waste. If it
has negative impact on development, then it is even worse than waste, it =
is actually
harmful.
Owen
