IPv6 Adoption Statistics: a Comparison of Different Metrics

In 2018, we compared the per country IPv6 adoption statistics as measured by Facebook and APNIC. In general, we found broad agreement, though notable differences did exist for some countries. Two years later, on the occasion of the RIPE NCC::Educa IPv6-only event, we're revisiting the topic and expanding the comparison to include results from measurements by Akamai, Google and Cisco.

The topic of measuring IPv6 adoption has gained momentum ever since IANA allocated the last availabe IPv4 /8s to the RIRs in 2011 and, subsequently, the IPv4 pools held by the RIRs dropped to levels where justified needs could no longer be met. For the world at large, in order to move away from IPv4 and switch to an IPv6-only public Internet, wide scale deployment is essential. Statistics which provide insights in how countries are progressing in this area are, thus, of interest to decision makers. Google started publishing a time series of the percentage of users reaching them over IPv6 more than a decade ago. They also provide a map-based visualisation which shows the latest levels of IPv6 adoption per country. APNIC, Akamai, Facebook and Cisco followed, each using their own, independent, methodologies, to compute and publish per country (and in the case of APNIC, with a breakdown per ASN) IPv6 adoption statistics. Though at first glance different teams trying to measure the same quantity may look like a duplication of efforts, it is actually good research practice. When all are in agreement, it boosts confidence in the result. When the numbers diverge, it highlights uncertainties, which can be taken into account when evaluating the results. All in all, parallel measurements help advance knowledge.

Unfortunately, all of the five organisations publish their results on separate web pages, in distinctly different colour schemes; that makes it difficult to compare the measured IPv6 adoption rates. To alleviate this, we took the raw data used for each of the published maps on 1 June 2020, merged these into a single table and created a new set of identically-styled maps, keyed by the respective organisation's metric.

Grouped in a single web page this allows for a quick visual comparison of the results. In addition, when hovering over the countries, the popups in each map display the metrics from all sources, which eases comparing the exact numbers. Figure 1 below shows a screenshot for Akamai, APNIC, Facebook and Google. We leave out Cisco at this stage because their IPv6 adoption metric is a composite, the result of combining a number of aspects of IPv6 deployment; these results are thus not directly comparable to results obtained by the others.

Figure 1: IPv6 adoption as measured by APNIC, Akamai, Facebook and Google respectively

At first glance, we already see how the United States and India have different intensities of green on each of the maps; a clear indication that the different methods calculated substantially different levels of IPv6 deployment for these two countries. When we turn to the raw data, we in general find reasonably good agreement; for most countries the percentages differ by no more than a few percent. We find the spread, the difference between the lowest and highest measured IPv6 adoption rate to be larger than 5%, for only 28 out of 150 countries. The following table lists them in decreasing order:
 

Table 1: Countries for which measured IPv6 adoption rates diverge by more than 5%

The largest differences are found for the small island nation of St. Martin, in the Caribbean. Akamai and APNIC measure well over 30% IPv6 adoption, while Facebook sees 3% and Google 0.0%. However, with a rather small population, this could be a bias that has been introduced by a relatively low number of measurements. In the APNIC data we see wide fluctuations in the daily measured percentages for St. Martin, over the past 12 months.

The next two in the table, United States and India, are more intriguing. Being among the most populated countries in the world, low numbers of measurements should not be the reason for the observed differences. APNIC's result of 68% for India, for example, is based on over 30 million samples. The fact that Google arrives at 23% lower IPv6 adoption for the country, suggests the respective measurements do not reach customers of the different service providers in the same proportion. Typically, not all ISPs in a country are at the same level of IPv6 deployment. Table 2 shows how, for India, APNIC measured 95% for Reliance Jio, around 60% for Airtel Mobile, Vodafone and Idea Cellular, but less than 1% for BSNL and Airtel Broadband. If, compared to APNIC and Facebook, Google sees fewer users from the mobile providers and more from the broadband providers, that would explain Google's significantly lower result for IPv6 adoption in India. Why this happens is something we can only speculate about. Without insights in the market share of each ISP in a country and per provider results from Akamai/Facebook/Google, it is also difficult to tell which of the four metrics is closest to reality when numbers diverge as much as in Table 1 above. There is no clear pattern either, for instance that one metric always provides the highest and another always the lowest score.

Table 2: IPv6 user capabilities of the top 10 ASNs in India, as measured by APNIC. 

As stated earlier, we left out results from Cisco in the above comparisons. The reason is that, unlike the others, Cisco's overall IPv6 deployment score is a composite metric, which not only looks at the percentage of IPv6 users in a country, but also take statistics on IPv6 transit ASNs,  as well as content reachable over IPv6 into account. The metric thus recognises that next to users, a country also needs (local) services to work with IPv6. The screenshot in Figure 2 gives an idea how the world is doing with respect to IPv6 under this metric. Compared to the user-based adoption statistics, many countries see their IPv6 adoption rates rise. However, no single country reaches 66% or more. And when ranking the different countries, we see the United States drop from ninth place in Google's IPv6 user adoption, to the nineteenth place in Cisco's overall IPv6 adoption stats. Apparently, other countries are doing a little better in nation-wide IPv6 infrastructure aspects.

Figure 2: IPv6 adoption map, keyed by Cisco's overall metric

 Conclusion

The per-country IPv6 user adoption statistics provided by Akamai, APNIC, Facebook and Google are in reasonable agreement for the large majority of cases. Differences of a few percent are within an acceptable range and do not really change the picture painted by these statistics. However, for a number of countries, the results diverge by much more than we would expect from pure statistical fluctuations. For some reason, the measurements by the different organisations appear to reach substantially different subsets of users. Which number is closest to reality for those countries, is hard for us to tell. We welcome input on that in the comment section below. Cisco's approach with a composite metric, taking a number of IPv6-related aspects into account is an interesting one. However, it also highlights the difficulties with summarising the complex issue of a country's readiness for IPv6 into a single number. One always needs to check the results from the different components to understand how the overall score of a country can be improved.