[arin-ppml] Draft Policy ARIN-2019-19: Require IPv6 Before Receiving Section 8 IPv4 Transfers

[Matthew Wilder]

Hi Michel,

I have observed the same trend over the years, and I completely agree that enterprise adoption lags. I think it can be explained a number of ways. In my experience wearing my IPv6 deployment hat at a major ISP, I know that we enabled IPv6 for our consumer subscribers who were not asking for it. At the same time, we made it available to enterprise subscribers, and there has been a (very) slow trickle of  enterprises adding IPv6 to their services. For a consumer to get IPv6 on their service requires no effort. For an enterprise to enable IPv6 on their network is some non-zero effort. That is perhaps the briefest summary explaining the delta.

I have a sense that the tension underlying this proposal exists between the economic cost of deploying, and the cost incurred as a result of those who are not deploying. Deployment of IPv6 requires effort - an economic hurdle high enough that some parties have not yet (or perhaps may never) rise over by natural means. On the other hand, those organizations who do not adopt IPv6 are beginning to create a cost to those who are adopting IPv6. As an example, I have on several occasions asked architects in my company to use IPv6 when designing private network infrastructure, and when they turn to vendors with this requirement, the response is that IPv6 is not supported, often not even on the roadmap. Here is a case where no eyeballs are involved (to your point) where IPv6 would be better, but where a limited demand for IPv6 support creates economic disadvantage for those who would otherwise prefer IPv6.

All that being said, the discussions I have seen so far are turning me off of this particular policy proposal. I am persuaded at this point that the most likely outcome of this policy is a gaming the system with vacuous "ceremonies" as others have suggested. I would be interested in hearing from others if my observation ring true regarding the cost that IPv6 non-adopters have on those who wish to deploy IPv6 more ubiquitously. Are others sensing this? Or is it strictly a philosophical ambition to move to a unified protocol version with no reliance on NAT?

Thanks!
Matthew


-----Original Message-----
From: Michel Py <michel at arneill-py.sacramento.ca.us> 
Sent: November 6, 2019 04:47 PM
To: Matthew Wilder <Matthew.Wilder at telus.com>
Cc: arin-ppml at arin.net
Subject: RE: [arin-ppml] Draft Policy ARIN-2019-19: Require IPv6 Before Receiving Section 8 IPv4 Transfers

Matthew,

> Matthew Wilder wrote :
> The Google IPv6 stats page clearly states that their graph indicates 
> the % of users who access Google services using IPv6. That means 
> eyeball networks, enterprise, non-profit, government, etc. In other words, you might summarize this by saying "the Internet".

No, you might not. Google does not measure IPv4-only traffic. Google measures the percentage of the people using it that are IPv6 enabled, which is not the entire Internet. It has nothing to do with the traffic on the Internet.

BTW, I have measured the IPv6 share on the vast majority of ISPs that are IPv4-only.
Guess what : it's 0%.

Although I will gladly agree that "everyone uses google" is close enough to be taken seriously, it measures only humans.
Hosts with no users do not use google. The Internet is not only humans. Some of us out there have computers on the Internet for different purposes than surfing the web.

And when you look at these 30% that are enabled, you will find that the lion's share of them is people who have whatsoever no control of their host : eyeballs with a cell phone completely locked by the carrier. What Google measures, mostly, is captive eyeballs. Not the Internet.

Case and point : open https://www.google.com/intl/en/ipv6/statistics.html
Zoom in so the entire chart shows the last year. You will see the weekly pattern.
Highs on Saturdays and Sundays, lows during the week. The only explanation is that people are using IPv6 more during the weekend because they are at home. This is confirmed by the sag that happens every year around NYE.
What does it mean : that enterprises are not IPv6 enabled.

This graph does _nothing_ to measure Internet traffic. You google, and you click on an IPv4 link. The actual traffic going through the Internet is IPv4, yet that graph says that you are using IPv6. It is not a valid measure of Internet traffic.

The Internet is not Google. The Internet is not IPv6-enabled captive cell phone eyeballs. Not only.
The valid measure of Internet traffic is at Internet eXchanges and on the backbone of ISPs.

IPv6 is deployed at 30% of people who use Google at home.
Less than 25% at the office, and this is very optimistic as some of the requests to google during week days would be made at home also.

The 1/6th reduction during weekdays means that, the 8 hours that users are sitting at their office desk, they use IPv4.

You are looking at the numbers that you want to see, not at the big picture.

One more time :

https://fedv6-deployment.antd.nist.gov/
Look at Industry IPv6 and University IPv6, the big tables at the end.
Maybe I'm color blind, but I see mostly red.

https://stats.ams-ix.net/sflow/ether_type.html
How do you explain that the traffic, which includes private connection between parties, at a major IX, is less than 3% ?
Because there is a cache ? Well if it's cached locally, it is not Internet traffic, or is it ?


Michel.