Six Years Since World IPv6 Launch: Entering the Majority Phases
As reported in an ISOC report last year, IPv6 adoption is now solidly in the "early majority phase" of the technology adoption life cycle by many metrics (well past "innovators" and "early adopters"), with progress beyond that in some areas. Akamai continues to see solid growth in IPv6, including significant traffic peaks during large events, and we continue to find cases where IPv6 outperforms IPv4.
Driven by the growth of internet end users outpacing the supply of IPv4 addresses needed to number their devices, and a desire to plan for a future IPv6-based internet, many networks around the globe are shifting to provisioning their end users with IPv6 connectivity. Currently, IPv6 deployment in 7 of the top 10 global economies has reached a point at whichAkamai typically sees more than 18% of requests to IPv6-enabled sites using IPv6, with India (56%), the United States (46%), Germany (35%), Japan (28%), and Brazil (27%) having much larger percentages. We have come a long way since the World IPv6 Launch when we reported seeing only approximately 1% of U.S. end users using IPv6. With about 80% of U.S. mobile end users accessing IPv6-enabled Akamai sites over IPv6 (and 90% from some carriers including Reliance Jio in India), the adoption of IPv6 in the mobile space is well through the "late majority" phase.
Over the past year IPv6 has been enabled on many of the largest online streaming events on the internet, as delivered by Akamai, such as numerous major sporting events including the VIVO IPL and some major broadcasters of the Olympics. During these events, we've often seen between roughly one-quarter and two-thirds of the global event traffic delivered over IPv6, depending on the audience distribution. We often see a considerably higher average throughput to IPv6 end users than to IPv4 end users during dual-stacked events (although some of this may be due to sample bias). We've even seen multiple recent cases where massive IPv4-only events caused problems for CGNAT middle-boxes but IPv6 traffic continued to flow unimpeded.
For dual-stacked hostnames we typically see higher average estimated throughput over IPv6 than over IPv4. Some of this may be due to IPv6-connected users being correlated with better connectivity, but more than half of dual-stacked hostnames (weighted by daily bytes delivered) have IPv6 estimated throughput at least 50% faster than IPv4, and 90% of these hostnames have the IPv6 estimated throughput at least 10% faster than IPv4. Individual sites and network providers will vary (with IPv6 performing worse than IPv4 in some cases). However, the overall trend aligns with past results from LinkedIn, Facebook, Akamai, and others that show IPv6 as delivering significantly better performance than IPv4, especially in some mobile networks.
IPv6 content availability on Akamai
Aggregate IPv6 traffic levels are a function of whether users have IPv6 connectivity through their network/ISP, whether content is available via IPv6 ("dual-stack IPv6+IPv4"), and whether client devices support IPv6.
For the past two years, Akamai has had IPv6 enabled by default for customers integrating new properties onto our platform (although we still need to support customers who opt out, such as because their backend origin or log processing infrastructure has problems handling IPv6 addresses). Over this time, we've seen significant growth in the percentage of dual-stacked customer hostnames on our platform, although there is still quite a way to go.
Fig. 1: Portion of hosts dual-stacked on Akamai by request volume
Over these two years, we have seen a doubling in the number of dual-stacked hostnames across a range of categories, including for sites with more than 100 million requests per day that include the websites of many household-name brands (Figure 1). Most of these categories are also now in the "early majority" adoption range. (Recent overall growth has been slower and more complicated at the top end of the spectrum, but that also has had a higher level of IPv6 adoption.) We continue to explore ways of safely accelerating the transition of customer content and sites to IPv6+IPv4 dual-stack, and customers are encouraged to dual-stack existing and new content.
Many customers see IPv6 as a strategic imperative and have been working hard to dual-stack content. For example, leading broadcaster NBC Universal sees the needs-based and value-based benefits of IPv6 and continues to be a leader in making dual-stacking part of their standard operating procedure.
Fig. 2: Distribution of "IPv6 usage" percentages (IPv6 requests / total requests) across request-weighted hostnames, showing the wide distribution of potential "IPv6 global user adoption" numbers for different types of content
Customers sometimes ask what percentage of requests to expect over IPv6, but there is no one single number. The "IPv6 usage" percentage can vary widely based on end-user population. As a result, global and country-level IPv6 measurements can be sensitive to the mix of content being measured and the associated end-user bases.
Looking at distribution of the "IPv6 usage" percentage across a 24-hour snapshot of traffic against around 10,000 dual-stacked customer hostnames each serving over 250,000 requests per day (weighted by total requests per hostname) in Figure 2, we see that most hostnames have IPv6 usage at or below 45%, but that more than three-quarters of (weighted) dual-stacked hostnames have IPv6 usage over 15%. (We've sometimes seen some large events with a heavy mobile user population in India reach over 65%.) Some of the modes include consistent classes of traffic: for example, the mode around 45% IPv6 usage contains lots of U.S.-centric streaming media content. Those modes near 0% often have end users in countries with little-to-no IPv6, use a client application lacking IPv6 support (such as some desktop apps and consumer electronics), or are web-service calls from cloud services lacking configured IPv6. Those with very high IPv6 usage are often from mobile apps in countries with high IPv6 mobile adoption.
End-User IPv6 Adoption Drivers and Ramifications
Akamai continues to see growth in IPv6 adoption in many networks, including mobile and broadband ISPs. The IPv6 growth we see in countries is a function of the IPv6 adoption within the networks deployed in that country.
Per some recent analysis by APNIC's Geoff Huston, there's no one clear factor yet as to which networks are and aren't deploying IPv6 and to what degree. However, freely available IPv4 addresses were exhausted by all but one of the RIRs prior to September 2015 (with Africa's AFRINIC projected to run out in less than a year). IPv4 address pricing on secondary auction markets has also been increasing, with many blocks selling in the range of $15 to $16 per address in 2018.
Between the cost of acquiring IPv4 addresses and the complexity of running a dual-stack network with both IPv6 and IPv4 as equivalents, it is becoming increasingly attractive for network operators to run "IPv4 as a service" (IPv4aaS) over IPv6. Technologies such as NAT64/464XLAT, lw4o6, MAP-T, MAP-E, DS-lite, and even HTTP proxies and VPNs can all leverage an IPv6-only network with IPv4aaS being used to provide end users access to IPv4-only resources. A crowd-sourced spreadsheet shows which technologies are being used by various ISPs.
Networks that have control over their end-user devices, such as mobile carriers and streaming set-top boxes (such as the Xfinity X1) can readily leverage IPv6-only deployment for those devices, while using IPv4aaS for any residual IPv4 traffic.
One ramification of IPv4aaS and common to all of the transition technologies is that IPv4 traffic relies on additional infrastructure with varying scaling and resource availability characteristics (such as limited NAT capacity). We've observed cases in which large online events or software bugs have caused problems with the NAT infrastructure relied on for IPv4 but the IPv6 has continued to work just fine. This is a compelling reason to dual-stack content: Even if your content isn't responsible for triggering such a problem, providing your end users with IPv6 as an option allows them to bypass these issues and achieve higher availability and/or performance in the face. Dual-stacking large events can also help reduce the chances of exhausting NAT resources within ISPs and causing problems for others.
Another ramification of IPv4aaS is that while IPv6 may break out onto the internet locally (such as in cities scattered around a country), we have started to see cases where IPv4 traffic is all passed through more centrally located CGNATs or other gateways. As such, a CDN is able to get closer to users over IPv6 than over IPv4 (without deploying behind the CGNAT), providing better performance and scale.
Growth in client addresses
In a single sample day in March 2018, we saw HTTP(S) requests from around 1.7 billion unique IPv6 client addresses (and approximately 10 billion native IPv6 client addresses in a March 2018 sample week). As IPv6 clients may change their addresses periodically (e.g., because of privacy addressing), this likely significantly overestimates the number of unique IPv6 endpoints. During the sample day in March 2018, we saw 576 million active /64 prefixes covering these IPv6 client addresses, with 1.43 billion of these prefixes in the March 2018 sample week. Compared with IPv4, we actually see more active /64 prefixes from HTTP(S) clients in a week than we see active IPv4 addresses in a year. This continues to provide evidence for the freedom-in-addressing that IPv6 offers.
Most of these have increased by approximately an order of magnitude over the past four years (Figure 3) — in March 2014 we saw 157 million /64 prefixes in a week from clients vs. 1.43 billion this March, with similar increases across the other metrics. Also, for comparison, we see more unique active IPv6 client addresses in a week than the total number of possible IPv4 addresses.
Fig. 3: Increase in active IPv6 client addresses and unique /64 prefixes making HTTP(S) requests to Akamai during a week, over the past four years (excluding Teredo and 6to4)
End-user adoption by country
Akamai's measurements of end-user IPv6 adoption see a few hundred billion HTTP(S) requests over a 24-hour period (a weekday such as a Wednesday for these measurements) across a mixture of hostnames and look at the percentage of requests over IPv6 relative to total requests. Depending on the geographic (and device) mix of traffic, we often see global averages for dual-stacked hostnames between 17% and 31% IPv6 usage, with some being well outside this range on either end (Figure 4). Even within a country, the IPv6 usage will vary based on the mix of mobile, broadband, and business traffic, as well as the targeted devices. As such, we'll see fluctuations in the country-level average as the mix of dual-stacked content we deliver changes over time or as we look at different sample sets of dual-stacked hostnames. However, the trends we see line up similarly to those reported by Facebook, APNIC, Google, and others, each of which reports by a different sample set.
We continue to see upward growth over the past five years in many countries. While some countries such as the United States, Japan, Belgium, Germany, Malaysia, and a few others had major networks start their IPv6 rollouts five years ago with steady growth over that time, India, Taiwan, and Mexico and other countries had networks start deploying IPv6 at scale more recently. As mentioned above, the United States, India, and Belgium all see approximately half of traffic to dual-stack sites delivered over IPv6 depending on the mixture of traffic.
Fig. 4: IPv6 usage/preference for select dual-stack hostnames on Akamai by country (measured on weekdays, which are typically lower than weekends)
End-user adoption by network
Many of the largest networks in the world (in terms of subscriber base) have been aggressively deploying IPv6 over the past few years, with a majority of requests using IPv6 to reach dual-stack sites in these networks. For example, across the top four U.S. mobile carriers we now see IPv6 usage rates of greater than 80% and this continues to steadily increase. Reliance Jio, the largest mobile carrier in India, has IPv6 usage of greater than 90% based on our observations. Some of the largest fixed-line networks in the world (Comcast, AT&T, Deutsche Telekom, British Sky Broadcasting, Rogers, KDDI, and others) all have IPv6 usage around or well above 50% in our observations, with many others on a path to reach and cross that soon (Figure 5).
Fig 5: IPv6 usage/preference for select dual-stack hostnames on Akamai by network (measured on weekdays, which are typically lower than weekends)
Residual IPv4
In addition to looking at IPv6 adoption, we've also started to look at "residual IPv4." By looking at areas with the highest levels of IPv4 traffic on dual-stack hostnames, we can identify areas that will have the biggest impact on global, country, or network IPv6 usage metrics. These tend to fall into the category of areas where there is already high IPv6 usage (but still significant opportunity for growth) and those areas with very limited IPv6 deployment.
Residual IPv4 by country
For example, we have seen that the following countries have high significant IPv6 deployment but also high residual IPv4: the United States, the United Kingdom, Japan, Germany, India, France, Canada, Brazil, Australia, Mexico, the Netherlands, and Sweden.
On the other hand, we've seen that some countries have very limited (less than 3%) IPv6 usage — Russia, China, Italy, Spain, Indonesia, Turkey, and South Korea — which means that most of their traffic is residual IPv4 traffic.
For China, in particular, although there are plans to deploy IPv6 broadly, we have not yet seen much IPv6 traffic materialize against dual-stacked sites outside of China from networks other than CERNET2. As seen in other countries, IPv6 deployment has the potential to ramp up quickly, especially in countries with a smaller number of ISPs.
Residual IPv4 by network
On the network side, more IPv6 movement will be needed in the longer tail (such as in smaller ISPs and corporate networks) to keep global IPv6 adoption moving upward. Smaller networks (with correspondingly less aggregate residual IPv4) are less likely to have deployed IPv6 than the larger networks. As of May 2018, we have observed that the top 60 networks constitute approximately 50% of the residual IPv4 traffic, with 58% of these having IPv6 usage rates of greater than 2%. Looking at the top 387 networks that constitute 80% of residual IPv4 traffic, this drops to 31% having IPv6 deployment of greater than 2%. The top 90% of residual IPv4 traffic comes from approximately 1300 networks and only 17% of them have IPv6 deployments of greater than 2%.
On an optimistic note, there are many network partners where Akamai has been able to obtain functioning IPv6 connectivity to our servers but where the network partner has not yet rolled out IPv6 to their end users. As of May 2018, we now have IPv6 live on Akamai servers in 114 countries, more than 700 cities, and over 850 network providers.
We also have seen continual growth in the number of origin ASNs from which we observe active IPv6 HTTP(S) clients arriving, from 3,842 in March 2014 to 6,887 in March 2018.
Residual IPv4 by client devices
On the device/client side, all recent major mobile and desktop operating systems and web browsers support and use IPv6, but they are not the only clients. We see very high levels of residual IPv4 traffic from some popular set-top boxes and gaming consoles that do not yet support IPv6. Additionally, some custom applications (such as download clients, games, and mobile apps) do not use IPv6 and thus can also result in residual IPv4 traffic. Consumer electronics manufacturers and app authors are strongly encouraged to support IPv6 in their client software. Using a scheme such as RFC8305 Happy Eyeballs Version 2 (as implemented in operating systems such as recent versions of macOS and iOS) can allow clients to also improve their reliability and performance to continue functioning even if either IPv4 or IPv6 breaks but the other continues to work.
Conclusion
Although we still have a way to go before residual users of IPv4 are truly "laggards," six years after the World IPv6 Launch we are at a stage where the use of IPv6 should be considered the norm and certainly included in default planning for any new projects. For content providers, ISPs, and device manufacturers who did not want to be early adopters and have not yet enabled IPv6, now is as good a time as any to take the plunge. Doing so may have performance and reliability benefits in addition to preparing you for the future.
For more measurements on IPv6 adoption, see:
Thank you to Dave Plonka, Moritz Steiner, and others for providing some of the details for this blog post.
While precautions have been taken in the preparation of this document, Akamai Technologies, Inc. assumes no responsibility for errors, omissions, or for damages resulting from the use of the information herein. The information herein is subject to change without notice. Akamai and the Akamai wave logo are registered trademarks or service marks in the United States (Reg. U.S. Pat. & Tm. Off). Non-Akamai product and corporate names are trademarks or registered trademarks of other companies; they are used only for identification purposes and to the owner's benefit, without intent to infringe. Published June 5, 2018.
