Apple: connections established 40% faster with IPv6 than with IPv4
Any delay opening a web page costs money
Any delay opening a web page costs money
Opinion has long been divided as to whether IPv6 is or isn't faster than IPv4. Some studies have found evidence that IPv6 is indeed faster, while others (including one focusing specifically on IPv6 and Facebook) have drawn the opposite conclusion. In many cases, comparative research has failed to detect any significant speed differential, but some investigators have suggested a trend, with IPv6 gradually catching up to IPv4 over several years. The various studies are summarised here [1, 2]. Viewing the evidence in the round, our impression has so far been that the two protocols are similar in terms of speed. However, at Apple's Worldwide Developers Conference (WWDC20) in June of this year, software engineer Jiten Mehta urged delegates to make use of the latest internet networking technologies – especially IPv6 – when developing apps. His contention was that, on average, IPv6 network connections are established 1.4 times faster. The main reason apparently being that native IPv6 connections don't use (CG)NAT and therefore route IPv6 traffic more efficiently.
In theory, IPv6 should be slightly faster than IPv4, since it has certain optimised header features, enabling IPv6 packets to be routed faster. What's more, (native) IPv6 connections don't require in-transit address translation of the kind needed with any IPv4 connection that involves (CG)NAT (i.e. most IPv4 connections). On the other hand, IPv6 support was added to network equipment later. That could mean that some ASIC hardware's IPv6 support isn't realised in an ideal way, and that router enhancements might tip the balance in favour of IPv4.
Stats published by APNIC show a clear improvement over time in IPv6's performance relative to IPv4. In recent years, the difference in average round-trip times (RTTs, or ping times) has steadily declined.
It is apparently only because of the situation in Asia that IPv6 has yet to overtake IPv4 in the global speed stats. "The countries whose performance is worst are countries with highly restrictive internet access policies (such as China, with its Great Firewall)," says Marco Davids, Research Engineer at SIDN Labs. "That may well influence speeds. Another possible reason why IPv6 doesn't outperform the older technology is that peering may not be as good for IPv6, meaning that network packets have to take a slower route to their ultimate destination."
If we look only at Europe and America, we see that IPv6 now outperforms IPv4. In the last of the three graphs below, you can see that the Netherlands is following a typical European development path.
Lee Howard, CEO of IPv6 access provider Retevia, has long argued that IPv6 is faster than IPv4. In a presentation [1, 2, 3] to the NANOG 76 conference last year, he showed how native (dual-stack) IPv6 and IPv4 connections are generally equally fast. On mobile networks, however, IPv6 connections over NAT64 appear to be up to 25 ms faster than IPv4 connections over NAT64. The most likely explanation for the discrepancy is the need for NAT translation or a delay in the CLAT software on mobile devices. Howard claims that IPv6 has the edge on landlines as well, with data indicating that the average IPv6 connection is realised 15 ms faster than with IPv4. Moreover, on landlines, the difference appears to depend on the IPv6 fail rate. According to Howard, the explanation may lie with the 'Happy Eyeballs' algorithm (RFC 8305), which enables dual-stack applications to initiate parallel IPv6 and IPv4 connections more or less simultaneously. As a result, the worst IPv6 scores aren't included in the stats, he suggests, because they result in a failure.
On the internet, speed matters. Delays that affect end users (web pages that are slow to load), have a direct impact on reach and earnings. Amazon once said that every 100 ms delay resulted in a 1 per cent loss of sales. According to Akamai, cumulative speed-related losses can be as high as 7 per cent. And Google discovered long ago that a half-second increase in the time it took for its home page to load translated into a 20 per cent drop-off in traffic, no less. Google also takes page loading time into account when ranking search results. More information about the impact of long loading times is available here [1, 2, 3].