[rpd] IPv6 as a criteria in IPv4 Soft Landing AFPUB-2026-v6-001-DRAFT01.

[Jaco Kroon]

Hi Jordi,

On 2026/07/08 13:08, jordi.palet--- via RPD wrote:
> Hi Jaco,
>
> Some responses below, in-line.
>
> Regards,
> Jordi
>
> @jordipalet
>
>> El 8 jul 2026, a las 9:01, Jaco Kroon via RPD <rpd at afrinic.net> escribió:
>>
>> Hi All,
>>
>> I'm looping back to this original email, not because I want to answer 
>> here directly, but rather because I think it's appropriate to 
>> consider the "bigger picture" of responses.
>>
>> First and foremost, I've seen quite strong opinions from newly 
>> registered @gmail.com addresses.  I'd like to understand the 
>> affiliation of these participants.
>>
>> In my opinion there are a number of conflicting and not quite obvious 
>> agendas in this, for this, and against this, policy proposal - all of 
>> them with various commercial impacts for various entities. Almost 
>> none of them to the benefit of the "end consumer".
>>
>> I personally believe the best thing is to retain the current status 
>> quo *as it stands*, as this leaves operational decisions to the 
>> various network operators rather than enforcing operational issues by 
>> way of policy.
>>
>> That said, I *personally* do not believe IPv4 should have a future of 
>> any kind.  Here I'm in agreement with Jordi that networks should be 
>> deploying as IPv6-first.
>>
>> The *reality* however is that the world is predominantly IPv4 - 
>> "everyone" is connected using IPv4, and I don't (at least not in 
>> Africa) see that this is going to change any time soon.  We've been 
>> debating the reasons for this in the office the last week or two, and 
>> I'm not going to bore you with the details, but our opinion is that 
>> this is not because networks and techies don't want to switch but 
>> because customers (yes, you read that right) demands that we don't.  
>> When you deploy DNS for a customer - just dare try and ask for AAAA 
>> details when only provided with A details and see the response you 
>> get.  Usually from people that should *know better*.
>>
> I disagree, not true anymore. Measurements from Google, Meta, APNIC, 
> Akamai, etc., show 50% and none of them is able to measure the traffic 
> in China (where they have strict mandates for IPv6-only), so 
> worldwide, we are closer to 65-70%.
>
> In addition to that, there is something that we never count on 
> measurements, in terms of actual terabits. All the traffic from CDNs 
> is measured only “once” (when it is cached), but if users have IPv6, 
> they will get as many copies with IPv6 as users accessing that 
> content. This is the reason why IX can’t measure CDN traffic and often 
> they are showing much lower real data.
None of which gives the registry the rights to dictate to net-ops. Nor 
were I referring to any of the names you're mentioning, and I 
specifically refrained from - and will continue to refrain from - naming 
names.

Your entire argument is valid, but doesn't address my statement that 
real-world customers (at least in Africa) absolutely resist IPv6. Now 
you're dictating to netops that they must dictate to their customers.  Y 
ou cannot do that.  That's not the purpose of a registry.  Nor should it 
be.
>>
>> When people run speedtest (which I understand measures only L7 
>> throughput) and they get a lower value on IPv6 than on IPv4 ... 
>> what's the next thing that happens?  User logs into router, find the 
>> "disable IPv6" button (or equivalent) *click*.
>>
> If you get lower measurements with IPv6 there is some deployment 
> issue, either in the access, distribution, BNGs, or even upstreams. 
> This is not only creating this “slower effect”, but also Happy 
> Eyeballs entering into action and falling back to IPv4. One of the 
> reasons that I mention in a previous email a few days ago which means 
> that instead of having 70-90% IPv6 traffic you may be getting only 30%.

Right.  Let's take a 1Gbps ethernet link.  Max sizes packets, no VLANs.

ipg = 96-bit times aka 12 bytes
preamble = 8
eth header = 14
payload = 1500 (ipv4=20, ipv6=40, tcp=20, so payload for v4 = 1460 for 
v6 1440, excluding crypto overhead which may apply).
trailer = 4 bytes.

That means at max packets we can use:

packets per second = 1E9 / ((12 + 8 + 14 + 1500 + 4) * 8)
   = 1E9 / (1538 * 8)
   = 81274.38231

Which means:

IPv4 rate = 81274.38231 * 1460 = 118660598.17260B/s = 949284785.38080bps 
= 949.28Mbps (which is what we're typically seeing on 1Gbps services 
through speedtest).
IPv6 rate = 81274.38231 * 1440 = 117035110.52640B/s = 936280884.21120bpx 
= 936.28Mbps (which a customer is saying Oh I'm getting lower 
performance on IPv6 compared to IPv4 .. *off*).

>
> Geoff Huston do lots of measurements that show that IPv6 typically is 
> faster (not because the protocol itself, but just because the lack of 
> NAT and CGN, as well as because often it is deployed much better than 
> keeping the older IPv4 configs).

This techs all over the world (myself included) will agree with you, but 
the only measure that matters to customers (and many an "IT 
Professional") is Ookla Speedtest.

Absolutely irrelevant.  You're still dictating network design and 
architecture to entities that's not you, something you have no right to do.

>
> Here is the thing. When you deploy IPv4, you really check everything, 
> ensure a good monitoring, etc., right? Well, we like it or not, I see 
> many IPv6 deployments really broken. Again lack of expertise in real 
> deployment and good knowledge of standards.

Well ... here I get where you're coming from, and I have to concede that 
- but this isn't the case on all networks.  Again, irrelevant to the 
discussion.

>> Okay, so follow the (apparent) European standard of users may not 
>> have access to their routers where they're forced to have IPv6 on and 
>> can do nothing about it ... had a discussion the other day with 
>> someone in Europe about this, "I get better throughput by putting my 
>> own router behind that of my ISP and running IPv4 only on my own 
>> network" - where there's v6 only ... well, I've yet to hear/see about 
>> that end-network.  Some core networks sure I've heard of that.
>>
> This is probably from someone misinformed. In fact, since many years 
> ago EU regulations enforce the provider to give complete access to the 
> CPE, and even be able to replace it with your own, etc.
Uhrm.  Again, can't disclose, but no.  Two cases.  The one I know from 
face to face direct, and can confirm that no access to the router was 
available.  Ports aren't even opened towards LAN side. The other case 
was from a reputable developer who works on one of the Linux distributions.
>>
>> As much as I hate to say this:  IPv4 is going to be with us for a 
>> while.  And we should encourage deployment of IPv6, but I don't think 
>> we can enforce that by way of policy as this is interfering with 
>> network operational decisions/designs.  I also strongly disagree with 
>> apparent stances I've observed hinting that we should make the soft 
>> landing more lax as that just drives up costs for everyone much 
>> sooner and faster, and in a world where everyone I communicate with 
>> are already complaining about the cost of stuff I don't think that's 
>> fair either.
>>
> So we prefer instead of our own policies at our own pace, government 
> regulations?

I'm just saying your way isn't in my opinion the way to get this done.  
You're adding significant additional operational issues to measure 
metrics that networks honestly shouldn't have to need to care about, 
burdening network operators that's already suffering from an 
over-abundance of stupid.

I'm also saying that going the other way and completely loosening the 
soft-landing is even more absurd as it would have the side effect of 
blocking new entrants into the market (bad for consumer) much sooner, 
and make entering into the market much more expensive for new players.

At best you can state that the network operator has to be able to show 
which proportion of their network is IPv6 capable, and what portion of 
customers has access to IPv6 if they so choose. You cannot dictate that 
networks has to force IPv6 onto their customers who don't want it, and 
you most certainly (at least in Africa) cannot measure uptake based on 
traffic percentage.  I've included the reasoning on this into a 
presentation earlier in the year, and have managed to consolidate the 
different measures to within roughly 1% to explain the discrepancies 
between a couple various measuring methodologies.

This would put my own network at 99.885 % measurement.  But if I use 
your measurements we're below 10% - not because we've not done our part, 
but because:

a)  Consumer networks accessing services hosted on our network is IPv4 only;
b)  Customers are actively switching OFF IPv6 on the services they 
procure from us; and
c)  Customers that host services on our network aren't deploying AAAA 
records, even when we push them to do so.  In many cases we don't even 
know what all the names are that's being set to the relevant IP space.

You are now aiming to punish our (and other networks in a similar 
situation to ours) because of other's behaviour.

If I need to choose between that and government regulation ... I think I 
might go fishing instead.

Kind regards,
Jaco
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.afrinic.net/pipermail/rpd/attachments/20260708/c78bf380/attachment-0001.html>