[address-policy-wg] 2017-03, New Policy Proposal (Reducing Initial IPv4 Allocation, aiming to preserve a minimum of IPv4 space)

Hi,

am 23.09.2017 um 23:51 schrieb Willy MANGA:
> Hi Nick,
>
> Le 23/09/2017 à 21:41, Nick Hilliard a écrit :
>> Willy MANGA wrote:
>>> being a newbie here can you please explain briefly why, as of today ,
>>> these people really need IPv4 addresses ?
>> I'd be tempted to answer, except that you sent this email from an ipv4
>> address.  So, please deconfigure all IPv4 addresses from your computer,
>> re-send the email, and then I'll answer.
> From where I stand (at home in Cameroon, Central Africa), My ISP has /32
> on v6 since 2013 but nothing deployed. Why ? I suspect ignorance and no
> incentive.
> I use to poke some friends working in telco fields; they say they are
> some work in the background :)
>
> From where I work, I did my best to deploy IPv6 in our networks [1] [2]
> Our ISP here has its v6 prefix since ... 11 years.  We are its first
> customer requesting v6 and they deploy it for us.
>
> I keep pushing in my area but as you may guess, it's the most difficult
> one but I succeed at least in my organisation.
>
> So again, why do they rely on v4 (only) ? I really want to understand
> hurdles on european continent.
>
> I hope this time, it will be clearer :)

Oh, the question was clear already; the answer was a bit of a puzzle.

There are two Internets these days, the really old, 32 bit one, and the other one, based on 128 bit addressing.

Different infrastructures, that can run alongside each other over the same cable, but don't know of the other.

As you've noticed yourself already, IPv6 adoption is slow. So there's less incentive to run services on IPv6, which increases the need for IPv4 addresses.

It's not too difficult to predict that IPv4-only services and servers will be around for the next decades.

As you already have IPv4 and IPv6 address space, you can run Dual Stack, e. g. have v4 (maybe NATed) and v6 on each client, so each of your client systems can reach v4 and v6 destinations.

If you don't have any publicly routed v4 address, you're be restricted to the upcoming, nice-and-shiny, IPv6-based, end-to-end Internet. Which, unfortunately, is only a subset of what is available in the rusty, IPv4-based, initial version.

I recently tried it myself, setting up a VM for an audience that "usually" is v6-enabled, so I thought "cut the crap, we do this one v6-only". So the VM only received v6 networking. And then I tried to update it to the current patchlevel — and I noticed that v4 is still a necessity:

root at discourse:~# host de.archive.ubuntu.com
de.archive.ubuntu.com is an alias for ubuntu.mirror.tudos.de.
ubuntu.mirror.tudos.de has address 141.76.1.208
ubuntu.mirror.tudos.de has address 141.30.62.22
ubuntu.mirror.tudos.de has address 141.76.1.204
ubuntu.mirror.tudos.de has address 141.30.62.23
ubuntu.mirror.tudos.de has address 141.76.1.200
root at discourse:~#

No v6 address, so no updates. (Sure, I could have used uk.archive.ubuntu.com, which is available on v6 as well, but you will not always find a replacement URL that works on v6-only.) So that VM now does have an RFC1918 IPv4 address that's NATed on the Hypervisor ...

To make a long story short: if you want to start a new internet service, you need connectivity to the legacy, IPv4-based, Internet. There are v4-only customers out there (e. g. business customers of Unitymedia (Liberty Global's German branch) only get one public v4 IP for NAT and cannot get v6 at all), and this is not going to change anytime soon, as you can see in your area as well. Actually, even some public clouds didn't provide v6 until recently. And, in some cases, at least for parts of their fleet confirmed they never will.

The question at hand is: how long should the RIPE Community keep the door open? IPv4 is already restricted to new LIRs, and LIRs are supposed to hand that space down to their customers. If your business needs new IPv4 space, these days you have to become a new LIR, which as of now costs you a 2000 EUR signup fee and a 1400 EUR yearly fee. Currently you'll get a /22, i. e. 4 /24, with no questions asked (and an /29 of v6, IIRC). And a(nother) vote in the RIPE NCC assembly ...

2017-03 would change this to "you get a /24" per new LIR, effectively raising the cost of an /24 from ~85 EUR/month to ~173 EUR/month (assuming 36 months to "earn back" the signup fee), but not enforcing IPv6 deployment in any way.

If we, as the RIPE Community, really want to preserve the scarse IPv4 ressources for "new entrants" to be able to deploy v6, more is needed than just shrinking the assignment: IPv4 *is* done.
The only use case RIPE NCC should assign new IPv4 address space for is for documented and needed v6 transitions services — and it has to prohibit transfers on this space, as any transfer would nullify the initial use case anyway, therefore opening the path to reclaim the assignment. And to make this setup last even longer, only a /28 should be assigned, as you don't need 256 IPs for a 6-to-4-gateway service anyway (do you, if so, why?). Get this aligned with ARIN and the other RIRs, it should be quite possible to get a bunch of /somethings routed at /28 level.

But 2017-03 simply raises the cost per /24, without enforcing v6 deployment in any way. It therefore shouldn't pass.

Regards,
-kai