Re: IPv6 Traffic Re: IPv6? Re: Where to Use 240/4 Re: 202401100645.AYC Re: IPv4 address block

[Owen DeLong via NANOG <nanog () nanog org>]

> On Jan 14, 2024, at 19:50, Abraham Y. Chen <aychen () avinta com> wrote:
>
> Hi, Ryan:
>
> 1)     " ... it accounts for 40% of the traffic at Google.   ":
>
>     Perhaps you were referring to the following?
>
>     https://www.google.com/intl/en/ipv6/statistics.html

> 2)    If so, your quotation is correct, except there are some hidden stories below the surface:
>
>     A.    When you Google for it with key words "IPv6 Traffic Google", the first hit shows "IPv6 Adoption" that lead 
> to the above. So, strictly speaking, it is not traffic data that you are looking at.

Correct, that graph shows fraction of google unique end points that have IPv6 capability. It does not reflect traffic 
at all.

>     B.    Above the actual graph, you will find statements, such as "  ...  the availability of IPv6 connectivity 
> among Google users. ...." So, legally, the graph is correct on its own right, but may not be exactly what you 
> thought. Reader be aware!

Correct… I do not know of a graph showing traffic as a percentage for google.

>     It implies that the graph the IPv6 capability (equipment readiness) of Google users, not necessarily the actual 
> traffic they generate. The two do not equate to each other.

No, it shows actual IPv6 reachable, not equipment capability. Likely there is some relatively close degree of 
correlation between fraction of users and fraction of traffic, but you are correct that they are independent numbers. 
It’s entirely possible, I suppose, that that 45% of endpoints  reachable via IPv6 represents 10% of Google traffic and 
doesn’t really use Google very much at all. OTOH, it’s equally likely that 45% of end points is actually responsible 
for 90% of Google traffic. I doubt that either of these extremes is likely, however.

In many ways, however, the fact that 45% of eyeball endpoints have IPv6 reachability is much more meaningful than 
whatever random fraction of traffic they happen to represent.

>  
> 3)    However, the above did seem to support what was generally said in the public. Until, we found an interesting 
> ongoing (the only one of such resource that is updated about every ten minutes) statistics by AMS-IX (AMSterdam 
> Internet eXchange) :
>
>     https://stats.ams-ix.net/sflow/ipv6.html   
>
>     https://stats.ams-ix.net/sflow/ether_type.html
> a
>     The second URL shows that IPv6 accounts for approximately 5.7% of the overall Internet traffic that AMS-IX sees 
> today. If one traces back through the archived data, the earlier numbers were even much lower. In fact, those graphs 
> looked meaningless, because there was hardly any visible trace colored for IPv6. This has been going on for at least 
> the last one decade. So, it could not be an error.

This isn’t a surprise since the vast majority of Google’s (and most other content providers) traffic is delivered via 
private network interconnect and not on public peering points.

> 4)    We contacted AMS-IX for a possible explanation of the obvious discrepancy. They politely referred us to our own 
> ISPs. This triggered our curiosity. We decided that we needed to find the full world-wide IPv6 traffic data.
>
> 5)    There was an annual world-wide Internet traffic statistics and forecast published by Cisco that stopped after 
> 2017 (see URL below to the last issue). We contacted Cisco in 2020 and got an eMail confirmation.
>
>     https://cloud.report/Resources/Whitepapers/eea79d9b-9fe3-4018-86c6-3d1df813d3b8_white-paper-c11-741490.pdf

If you dig deeper on that, you’ll find that their data is purely estimated based on very limited collection.

> 6)    However, there has never been any equivalent publication for the IPv6 by itself that we could locate.

There is an interesting bit of data from Akamai in this post:
https://www.akamai.com/blog/trends/10-years-since-world-ipv6-launch#:~:text=Akamai%27s%20IPv6%20traffic%20levels%20and%20client%20base&text=As%20of%20May%202022%2C%20Akamai%27s,years%20ago%20in%20February%202020.

Which reports that 2022 Akamai IPv6 traffic was over 41Tbps, up from just over 1Gbps in 2012.

While IPv4 has grown in that same 10 years, I doubt that it has grown 4,100,000% in that same 10 years.

> 7)    In search for a possible explanation of the discrepancy between Pts. 1) & 3), we came across the following 
> article. In brief, it reported that the Peering agreements among Internet backbone providers were less settled for 
> IPv6 than IPv4. Thus, higher percentage of IPv6 traffic than that of IPv4 should have been directed through the IXs 
> (Internet eXchanges), such as AMS-IX.
>
>     https://www.theregister.com/2018/08/28/ipv6_peering_squabbles/

1. This is largely untrue today. Most IPv6 capable networks that peer on a public exchange with another IPv6 capable 
network set up sessions for v4 and v6 at the same time.

2. There’s a much more plausible explanation… Most of the big eyeball networks and most of the big content providers 
don’t deliver much of their traffic via public exchanges, yet they are the ones most likely to have IPv6 capability. 
While Akamai, for example, delivers a lot of traffic over Ams-IX, it’s mostly not to major eyeball networks which 
instead connect to Akamai over private peering. This artificially suppresses the IX perspective on the fraction of 
traffic that is IPv6 overall.

> 8)    The conclusion of Pt. 7) furthered our puzzlement, because it was opposite to what we were hoping for. That is, 
> the roughly 5.7% IPv6 traffic that AMS-IX sees implies that within the overall Internet, the IPv6 traffic should be 
> even less than 5.7%, not as high as Google's 40+% (Adoption) rate. Since we did not have the resources to further the 
> research on this topic, we saved the above summary to share with anyone interested in pursuing for a better 
> understanding. It will be much appreciated, if you could share your insights of this topic.

Well, the good news (see my second point in response to 7) is that it’s likely much larger than what you see on the 
exchange points, as you hoped.

According to CloudFlare in this article: https://www.theregister.com/2023/12/13/cloudflare_internet_traffic_2023/

1.      1/3rd (33.75%) of all internet traffic is IPv6.
2.      1/3rd of requests that could traverse IPv6 are being served on IPv4.

If both of those statements are true, then it indicates (theoretically) that turning off IPv4 tomorrow would be nearly 
lossless.
(33.75% * 3 = 101.25%).

I tend to suspect that the numbers might be a little off here, but the point remains that the amount of the internet 
that doesn’t work on IPv4 continues to shrink and at some point, the benefit of maintaining v4 will drop below its cost.

The sooner we reach that point, the less pain everyone will have to endure between now and getting there.

Owen