Where to buy Internet IP addresses

Joe Greco wrote:
> One of the goals of providing larger address spaces was to reduce (and
> hopefully eliminate) the need to burn forwarding table entries where
> doing so isn't strictly necessary. When we forget this, it leads us
> to the same sorts of disasters that we currently have in v4.

And if you are encouraging /48 handouts, /32 isnt large enough to
prevent that on the global level.

I don't know that I'm *en*couraging /48 handouts, but on the other hand,
I'm not sure I'm *dis*couraging it either.

On one hand, there's a reasonable argument to be made that the average
home user does not currently have enough devices to fill more than a
/124's worth of space.

But.

You have RFC3041 and similar techniques, stateless autoconfig, and a
variety of other general things that make it really awful for the default
ethernet network size to be something besides a /64.

Further, it seems clear from most discussions I've had, that people
really do want or need the ability to have multiple networks, for a
variety of practical reasons. Many of these have to do with keeping
different zones firewalled in particular ways,

So, really, I think the question is, how many unique firewalling
policies is a household likely to have, and then, maybe how many other
neighbors/friends/etc are also freeloading on that connection, each
with the same needs?

A /56 allows up to 256 networks. For today, that's pretty clearly
all that I can reasonably imagine even a sophisticated home network
along with several neighbors needing. Probably even within the next
ten years. At some point, however, it is possible that a /48 would
be a better choice.

I would definitely prefer to see a /56, or maybe a /48, handed out
today.

If we get into the practice of handing out /64's, it is just going
to encourage bad hacky design compromises and CPE/SOHO gear kludges
in the future?

... JG

You have RFC3041 and similar techniques, stateless autoconfig, and a
variety of other general things that make it really awful for the default
ethernet network size to be something besides a /64.

...

I would definitely prefer to see a /56, or maybe a /48, handed out
today.

When I first started looking at IPv6, the bottom 64 bits were
divided into the bottom 48 bits for a MAC address
and 16 more bits that could either be zeros or could be
used as a subnet number in roughly Novell Netware style
(modulo a local/global bit if you needed one).
If you wanted to assign addresses instead of autoconfiguring,
that was ok too, but it was obvious how autoconfig would work.
It was simple, clean, and flexible, and obviously intended that
an ISP would hand most customers a /64, which could easily
handle an entire building or medium-complexity campus.
Then I ignored those bits for a decade or more,
because nobody's IPv6 was much more than experimental.

When I came back, I found this ugly EUI-64 thing instead,
so not only was autoconfiguration much uglier,
but you needed a /56 instead of a /64 if you were going to subnet.
Does anybody know why anybody thought it was a good idea
to put the extra bits in the middle, or for IPv6 to adopt them?

Bill Stewart wrote:

When I came back, I found this ugly EUI-64 thing instead, so not only was autoconfiguration much uglier, but you needed a /56 instead of a /64 if you were going to subnet.

It's supposed to be a /48 per customer, on the assumption that 16 bits of subnet information is sufficient for virtually anyone; exceptions should be rare enough that they can be handled as special cases.

The /56 monstrosity came about because a US cable company wanted to assign a prefix to every home they passed, regardless of whether it contained a customer, so that they'd never need to renumber anything ever again. However, that would require they get more than the /32 minimum allocation, and ARIN policy doesn't allow _potential_ customers as a justification for getting a larger allocation, so they had to shrink the per-customer prefix down to a /56 to fit them all into a single /32. If all those assignments were to _real_ customers, they could have gotten a /24 and given each customer a /48 as expected. And, after that, many folks who can't wrap their heads around the size of the IPv6 address space appear to be obsessed with doing the same in other cases where even that weak justification doesn't apply...

Does anybody know why anybody thought it was a good idea to put the extra bits in the middle, or for IPv6 to adopt them?
  
Why the switch from EUI-48 to EUI-64? Someone in the IEEE got worried about running short of MAC (er, EUI-48) addresses at some point in the future, so they inserted 16 bits in the middle (after the OUI) to form an EUI-64 and are now "discouraging" new uses of EUI-48. The IETF decided to follow the IEEE's guidance and switch IPv6 autoconfig from EUI-48 to EUI-64, but FireWire is the only significant user of EUI-64 addresses to date; if you're using a link layer with EUI-48 addresses (e.g. Ethernet), an extra 16 bits (FFFE) get stuffed in the middle to transform it into the EUI-64 that IPv6 expects.

S

"64bit MAC" -- which pretty much exists nowhere. It's a repeat of the mistakes from IPv4's early days: CLASSFUL ROUTING.

I'm with you. I wish vendors and spec designers would just get over it and let people subnet however they want. If I want to set a network to be /96 or /120, I should be allowed to do so. Yes, I know autoconfig will not work -- and I don't want it to. I can make /31 IPv4 routes -- no router I've ever used complained about it. (that sends 2 addresses to one place; what happens in the place is not the router's concern.)

This has been a fascinating theoritcal discussion.. how do existing providers hand out space?

Hurricane electric (via its tunnel service) hands out a /64 by default and a /48 is a click away.

How do other providers handle it? I'm in the us and only have native v4 connectivity :frowning:

Do the various traditional last mile providers (sprint/Verizon/att/patch etc ) offer it for t1 and better? If they do then what do they hand out by default, what's available, at what price point and what's the upgrade path? Is it one click like he?

No provider I have talked to offers it for residential connectivity in the united states.
What does free.fr do?

If there is this level of confusion and disagreement around addressing schemes then will it ever be offered to residences over traditional last mile loops?

Ricky Beam wrote:

"64bit MAC" -- which pretty much exists nowhere. It's a repeat of the mistakes from IPv4's early days: CLASSFUL ROUTING.

Given there is no CLASS, but just a separation of network and host, I'd hate to compare it to classful routing. They probably would have been happy with a /96 network except for stateless autoconfig, which is quite nice for some stuff actually.

I'm with you. I wish vendors and spec designers would just get over it and let people subnet however they want. If I want to set a network to be /96 or /120, I should be allowed to do so. Yes, I know autoconfig will not work -- and I don't want it to. I can make /31 IPv4 routes -- no router I've ever used complained about it. (that sends 2 addresses to one place; what happens in the place is not the router's concern.)

I've not tried every vendor out there, but I've noticed some implementations handle /127 just fine from a routing perspective. I personally enjoy my /64 of /128 loopbacks. I'll be dead before I run out. :slight_smile:

Jack

Given there is no CLASS, but just a separation of network and host, I'd hate to compare it to classful routing. They probably would have been happy with a /96 network except for stateless autoconfig, which is quite nice for some stuff actually.

Ok, calling it "classful routing" might be a little melodramatic.

I would love to be able to set interfaces on my cisco hardware to /96's, but it's not allowed. Autoconfig screws that up. Even if it's not used, you're forced to live with it. Linux, BSD, etc. don't care. (and the instant they do, I can remove that stupid code.)

I've not tried every vendor out there, but I've noticed some implementations handle /127 just fine from a routing perspective.

So far, Cisco's gear is the only IPv6 routers I've messed with. And they will not let you set an interface to anything smaller than a /64. Loopbacks have slightly different rules, but in my case (IPv6 tunnels) that fact hasn't proven very useful.

In a message written on Mon, May 04, 2009 at 06:38:13PM -0400, Ricky Beam wrote:

So far, Cisco's gear is the only IPv6 routers I've messed with. And they
will not let you set an interface to anything smaller than a /64.
Loopbacks have slightly different rules, but in my case (IPv6 tunnels)
that fact hasn't proven very useful.

My 12.0(S), 12.4, and IOS-XR boxes are operating quite well with
/112's and /127's on GigE interfaces to each other, on GSR's, 7300's,
and 7200's. We also use /128's on loopbacks.

Must have been old (very old?) code when I first tried this. 12.4(15)T doesn't seem to care.

But it still doesn't allow transitional addresses:
gw(config-if)#ipv6 address 0::101:101/96
%FastEthernet0/0: Error: ::1.1.1.1/96 is invalid

gw(config-if)#ipv6 address 0::101:101/128
%FastEthernet0/0: Error: ::1.1.1.1/128 is invalid

If I cannot set an IPv4 address like that, the tunnel won't work. DHCPv6 requests end up outside the tunnel. (it's been many months since I messed with it, so the interface isn't in the config anymore, but the IPSEC setup still is :-)) I can send you the packet dumps if you want to scratch your head over it.

From: Ricky Beam [mailto:jfbeam@gmail.com]
Sent: Monday, May 04, 2009 6:38 PM
To: nanog list
Subject: Re: Where to buy Internet IP addresses

Given there is no CLASS, but just a separation of network and host,
I'd hate to compare it to classful routing. They probably would have
been happy with a /96 network except for stateless autoconfig, which
is quite nice for some stuff actually.

Ok, calling it "classful routing" might be a little melodramatic.

I would love to be able to set interfaces on my cisco hardware to /96's,

but

it's not allowed. Autoconfig screws that up. Even if it's not used,

you're

forced to live with it. Linux, BSD, etc. don't care. (and the instant they

do,

I can remove that stupid code.)

Actually, they will - you can set them to any arbitrary value between 1 and
128 (inclusive), atleast on every piece of gear I've tried to use.

I've not tried every vendor out there, but I've noticed some
implementations handle /127 just fine from a routing perspective.

So far, Cisco's gear is the only IPv6 routers I've messed with. And they
will not let you set an interface to anything smaller than a /64.
Loopbacks have slightly different rules, but in my case (IPv6 tunnels) that
fact hasn't proven very useful.

See above.
Maybe true on some gear, or some old IOS, or when using the auto-config key
word ... ?

/TJ

From: Ricky Beam [mailto:jfbeam@gmail.com]
Sent: Monday, May 04, 2009 7:23 PM
To: Leo Bicknell; nanog list
Subject: Re: Where to buy Internet IP addresses

My 12.0(S), 12.4, and IOS-XR boxes are operating quite well with
/112's and /127's on GigE interfaces to each other, on GSR's, 7300's,
and 7200's. We also use /128's on loopbacks.

Must have been old (very old?) code when I first tried this. 12.4(15)T

doesn't

seem to care.

But it still doesn't allow transitional addresses:
gw(config-if)#ipv6 address 0::101:101/96
%FastEthernet0/0: Error: ::1.1.1.1/96 is invalid

gw(config-if)#ipv6 address 0::101:101/128
%FastEthernet0/0: Error: ::1.1.1.1/128 is invalid

FWIW - ::/96 based addresses ave been deprecated for quite a few years.

/TJ

"64bit MAC" -- which pretty much exists nowhere. It's a repeat of the
mistakes from IPv4's early days: CLASSFUL ROUTING.

I'm with you. I wish vendors and spec designers would just get over it
and let people subnet however they want.

you can. there was a bit of a war in the ietf some years back, and yhe
64 bit boundary is a convention. hardware must route and forward on 128
bits.

do other than 64 and you do not get auto-conf. some do not consider
this a loss, others do.

So far, Cisco's gear is the only IPv6 routers I've messed with. And
they will not let you set an interface to anything smaller than a /64.

i wonder what strange gear you tried. all routers, cisco and other, i
play with operate on 128 bits.

randy

This is an important distinction.

- you CAN subnet however you like, with any number of bits in
  your prefixes

- autoconfiguration will work only in subnets with a 64 bit prefix.

The two matters are quite independent of each other, as far as I can
tell.

Regards, K.

Yesterday, it was.

You might want to read up about IEEE 802.15.4 and 6LoWPAN.
We are not joking when we talk about the next billion nodes on the Internet.

For those who are worried about running out on /56s:

There are 9000 trillion of those in the current 2000::/3 that is being handed out.
Each /56 is about a customer relationship, a home, a wire, ...
Say, each of them only needs a single dollar to get set up.
We are talking about 9000 trillion dollars being spent before we have to open up the next /3.
(2008's world domestic product, measured in purchasing power parity was about $ 69.49 trillion, by the way.
I'm talking about spending 129.6 years of world productivity for one dollar per /56, here.)

Folks:
There will *never* be a reason to hand out /60s, /62s, /64s, or, heaven forbid, /96s.
And I mean *never*:
It's much more useful to discuss this issue on fundamentals than on past practices.

Really, /56 for everyone is the only way back to an Internet.

Gruesse, Carsten

When I came back, I found this ugly EUI-64 thing instead,
so not only was autoconfiguration much uglier,
but you needed a /56 instead of a /64 if you were going to subnet.
Does anybody know why anybody thought it was a good idea
to put the extra bits in the middle, or for IPv6 to adopt them?

"64bit MAC" -- which pretty much exists nowhere. It's a repeat of the mistakes from IPv4's early days: CLASSFUL ROUTING.

Blame IEEE. They claimed that for identifying network cards 64 bit ID will be used in th future.... (Already used in IEEE 1394)

I'm with you. I wish vendors and spec designers would just get over it and let people subnet however they want. If I want to set a network to be /96 or /120, I should be allowed to do so. Yes, I know autoconfig will not work -- and I don't want it to. I can make /31 IPv4 routes -- no router I've ever used complained about it. (that sends 2 addresses to one place; what happens in the place is not the router's concern.)

I did not get any problem setting up any sunet length manually all the systems I tested.

Best Regards,
     Janos Mohacsi

Mohacsi Janos wrote:

When I came back, I found this ugly EUI-64 thing instead,
so not only was autoconfiguration much uglier,
but you needed a /56 instead of a /64 if you were going to subnet.
Does anybody know why anybody thought it was a good idea
to put the extra bits in the middle, or for IPv6 to adopt them?

"64bit MAC" -- which pretty much exists nowhere. It's a repeat of the mistakes from IPv4's early days: CLASSFUL ROUTING.

Blame IEEE. They claimed that for identifying network cards 64 bit ID will be used in th future.... (Already used in IEEE 1394)

Even if that is true and doesnt require ethernet as we know it to be forklifted which may be enough to rule it out from ever happening (see 1500 mtu for reference), address auto configuration does not require 64 bits. Its just nicer that way.

What IEEE is concerned about is global uniqueness purity.

Global uniqueness into perpetuity isnt required operationally on a lan, its just nicer that way.

Gateway directed auto-conf enable-able on any bit length, with icmp conflict detection seems no worse that what we have now with either dhcpv4 or apipa.

Joe

Older protocols, like classful IPv4, Appletalk etc. put a hard boundary
between the network and node portion. That was simple and, in the case
of IPX, Appletalk and DECNET, it was very convenient to have fixed
length network and node portions. IPv4 originally had a single boundary
between the network and node portion - if you look up the early
RFCs/IENs, the IPv4 addressing format was similar to Class A.

Of course in the case of IPv4, those classful hard boundaries caused
problems when we needed to squeeze more addresses out of the 32 bits by
moving to a fully varying boundary between the network and node
portions. IPv4 software in all nodes needed to be upgraded to work.

I think of the way IPv6 has done it is the middle ground. For
forwarding, the boundary between the network and node portions isn't
hard - it's purely longest match on the whole 128 bits. However, because
we've got so many bits, within a portion of the address space, a harder
(but not hard) boundary exists, to benefit from the convenience of
having fixed length node addresses, which results in things such as much
simpler autoconfiguration etc.

Regards,
Mark.

This has been a fascinating theoritcal discussion.. how do existing providers hand out space?

Hurricane electric (via its tunnel service) hands out a /64 by default and a /48 is a click away.

How do other providers handle it? I'm in the us and only have native v4 connectivity :frowning:

Do the various traditional last mile providers (sprint/Verizon/att/patch etc ) offer it for t1 and better? If they do then what do they hand out by default, what's available, at what price point and what's the upgrade path? Is it one click like he?

No provider I have talked to offers it for residential connectivity in the united states.
What does free.fr do?

Free does 6rd and allocate a /64 per customer.
Here is a presentation how they do this :
http://www.ripe.net/ripe/meetings/ripe-58/content/presentations/ipv6-free.pdf

Stephen Sprunk <stephen@sprunk.org> writes:

FireWire is the only significant user of EUI-64 addresses to date;
if you're using a link layer with EUI-48 addresses

Zigbee has been around a lot less time than FireWire, but is hardly
insignificant (ask anyone who's working on smartgrid or green home
stuff). I'm sure there are others that just aren't on any of our
personal radars.

-r