Exploring if IPv6 is actually faster across the Internet than IPv4
NANOG Presentations on IPv6 Performance
One way to explore how the Internet functions is to ask those service providers that actually provide core Internet network infrastructure. A good place to start investigating IPv6 adoption is to ask those Internet service providers that form the Internet backbone and connect people to it. The North American Network Operators Group (NANOG) is just such an organization that brings together those ISPs that construct the Internet.
At the recent NANOG 66 meeting in San Diego (February 8-10, 2016) there was an intriguingly-titled presentation “Is IPv6 Really Faster?” by Geoff Huston of APNIC. Geoff showed his gathered measurements obtained through providing IP-protocol-specific ads and by analyzing the resulting statistics. His presentation showed how 6to4 and Teredo contributed to most of the IPv6 connection failures. His statistics showed that IPv6 still has more connection failures than IPv4. His colorful graphs showed that IPv6 was slower than IPv4 back in 2012, but more current data shows that IPv6 is improving and can be equal to IPv4, and may even be faster than IPv4 in some cases. Geoff Huston was also interviewed on the topic of “Is IPv6 faster than IPv4?” where he discussed how IPv4 and IPv6 are very equivalent. Below is a picture from his presentation comparing native IPv6 (blue line) to 6to4 tunneled traffic (green line).
Geoff Huston also wrote an article titled “Examining IPv6 Performance” on November 26, 2015. Similar to his NANOG 66 presentation, he noted that one of the main issues that negatively affects IPv6 performance is the remaining systems that are still using 6to4 and Teredo tunnels.
Along these lines, there have been discussions at the IETF and in other forums to deprecate 6to4 and Teredo tunneling mechanisms. Almost a year ago, the IETF published RFC 7526, “Deprecating the Anycast Prefix for 6to4 Relay Routers.” However, Microsoft still has plans to use Teredo for Xbox, although they may be deprecating their Teredo servers for Windows systems.
At the NANOG 64 event in San Francisco, CA June 1-3, 2015, there was a presentation titled “Latency IPv4 vs IPv6: Understanding the difference” by Alexander Azimov, Qrator Labs. Alexander’s presentation started off by referencing a 2010 paper written by the Google team of Lorenzo Colitti, Steinar H. Gunderson, Erik Kline, and Tiziana Refice, titled “Evaluating IPv6 Adoption in the Internet” that was the first evidence that native IPv6 connectivity might be faster than IPv4 is some situations. Alexander’s presentation analyzed the BGP ASNs of service providers and discovered a disparity between Tier 1 providers for IPv4 and IPv6.
Also, at NANOG 64, there was a presentation titled “The benefits of deploying IPv6 only” by John Jason Brzozowski of Comcast, Geoff Huston of APNIC, Gaurav Madan of T-Mobile, and Paul Saab of Facebook. This presentation went over the IPv6 deployments of these service providers and one of the benefits each of the presenters spoke about was that IPv6 performs better than IPv4. During this presentation Paul Saab of Facebook showed that with their Mobile Proxygen code that they can control and measure traffic over one specific mobile carrier and discovered that native IPv6 was 40% faster than IPv4.
Facebook’s Experiences
For many years, Facebook has been a proponent of IPv6. Facebook participated in the World IPv6 Day event in 2011 and the World IPv6 Launch events. Their goals to build an IPv6-only infrastructure have been widely publicized. To Facebook, Internet performance is essential to making their user’s application experience the best it can be. Facebook prefers IPv6 transport for users reaching their platform because they believe that it is faster.
Paul Saab of Facebook has been involved in deploying IPv6 in their infrastructure for many years and has given many presentations on their experiences. Paul gave a presentation at the V6 World Congress 2015 event in Paris March 17-18, 2015. His presentation covered “Facebook IPv6 Strategy” and he mentioned that IPv6 is notably faster for Facebook users. Paul was quoted as saying “Facebook says it has seen users’ News Feeds loading 20 percent to 40 percent faster on mobile device…”. Below is a graph that shows that IPv6 is better than IPv4 from Facebook’s perspective.
Paul mentioned that one key problem limiting IPv6 adoption is the way that Java code is frequently written to use IPv4 and not use IPv6. This issue has been well-documented and there are ways to write effective dual-protocol Java code. But education is required for developers so they do not inadvertently disable IPv6.
Paul Saab also gave a presentation @Scale Conference in September 2015 titled “Facebook: IPv6 is Here and You’re Hurting your Users” or “Embracing IPv6 for Optimal Performance” where he stated that IPv6 was 15% faster than IPv4. At that time, Paul noted that over 10% worldwide Facebook users access their platform over IPv6, and over 20% of the Facebook users in the United States use IPv6 to reach their platform. It was also mentioned by Samir Vaidya, Director of device technology for Verizon, that 80% of Verizon’s mobile user traffic to Facebook now uses IPv6.
From these presentations, it is conceivable that if IPv6 provides better performance for Facebook then IPv6 may do the same for any other company (and their application or services).
HE IPv6 Deployment Progress
Hurricane Electric (HE) continually updates its Global IPv6 Deployment Progress Report with valuable data points about IPv4 and IPv6 comparing the two. There is a section of this report titled “IPv6 Performance” which reviews performance of DNS servers. Here is an excerpt from a recent report:
“Actually, the reason we looked at IPv6 reverse DNS servers was to get some hosts on both IPv6 and IPv4 so that we can compare IPv6 vs IPv4 latency.
IPv6 rDNS Nameservers where IPv6 is faster than IPv4 (by more than 1ms): 1731
IPv6 rDNS Nameservers where IPv4 and IPv6 are the same speed (within 1ms): 1456
IPv6 rDNS Nameservers where IPv4 is faster than IPv6 (by more than 1ms): 1004
IPv6 rDNS nameservers where IPv6 is as fast or faster than IPv4 (within 1ms): 3187
Percentage of IPv6 rDNS Nameservers where IPv6 is as fast or faster than IPv4 (within 1ms): 76.0%”
We can see from this report that IPv6 can be faster than IPv4 in at least three-fourths of the cases where both protocols are active.
Conclusions
Clearly, it is too early to declare that IPv6 always provides better performance than IPv4. IPv6 is still building momentum, but all the adoption graphs are displaying the “hockey-stick” property of rapidly going up and to the right. Today, depending on where you take your measurement, IPv6 can represent 10% to 20% of the total network traffic volume. However, it may still be a few years until IPv6 is 50% of the Internet’s traffic. Remember, that for every Facebook feed that loads over IPv6, it is one less Facebook feed that loaded over IPv4.
The IETF continues to create RFCs that refine and extend the IPv6 protocol. IPv6 will never be finalized, just as IPv4 continues to be a work-in-progress. The IETF may be taking steps to deprecate 6to4 and Teredo and vendors could soon remove these services from their end nodes. Geoff Huston wrote a recent article titled “Declaring IPv6 an Internet Standard” that discusses how there is still work to be done on IPv6 and how the IETF should finally fully commit to IPv6.
In the second part of this blog, we will explore the characteristics of IPv6 that have the potential to make it faster than IPv4. We will also review how you can test if your performance over IPv4 and IPv6 vary and confirm which one may be faster.
