Geoff Huston at APNIC has been doing a phenomenal job analyzing the performance of IPv6 vs IPv4. The way he does it is brilliant. You include an ad in webpages, an ad that is but a Java script that tests performance to a sink over IPv4 and over IPv6. And as long as you DO NOT click on that ad, Geoff doesn’t have to pay for it.
Geoff opened NANOG with a presentation where he shared his findings out of more than 20million measurements performed between September 2015 and January 2016. He looked for two things:
- How many IPv6 TCP connect sessions fail to set up (this number used to be pretty high in the past)
- Round Trip Time for IPv4 vs IPv6
His findings are both interesting and encouraging. The number of failed connect sessions decreased over time yet the problem still exists. These failures are also more prevalent in some Autonomous Systems than others. Her is a wall of shame:
How do we fix it? More monitoring, fixing the infrastructure, managing MTU and most importantly, getting rid of some of the tunnels that still provide IPv6 connectivity instead of native forwarding.
Second, Geoff saw good improvement in RTT for IPv6, this again due in part to the decrease in the use of tunnels such as 6to4. Now, IPv6 appears to be just as fast as IPv4. Here is his histogram for unicast traffic.
We too see parity in many cases as well however, we can clearly detect the significant impact that some of the structural IPv6 tunnels have on performance as well. See more on our earlier blog //www.nephos6.com/not-all-sps-are-created-ipv6-equal/
Geoff will continue his good work and we should all follow it but for now, his message was: “Don’t you tunnels just because you can, use them because you have to”.
I add to that: “Keep monitoring, there are more and more people roaming the IPv6 forest and they will hear the trees falling”.