Is IPv6 Faster Than IPv4?

TL;DR: IPv6 is faster than IPv4 ~39% of the time (locally anyway!).

To answer this question we decided to look at anonymized data we had collected from a range of monitoring agents over a 30 day period. For simplicity, the data used only focused on the client’s RTT (Round Trip Time) to their default gateway. This sub-path does not cover a full round trip to Internet based assets, it’s just the first leg of a journey (often over Wi-Fi), which admittedly can suffer extreme variability. There are many nuances that may affect our initial results, including but not limited to, size of data set, differences in device, OS (Operating System) level, the medium, protocol, CoS type, etc. It is, however, precisely because of the complexity involved that we decided to look at one simple and comparable metric!

The Approach

We began by using our Wi-Fi assurance and remote troubleshooting tool called PanSift (see a live demo). We started with 2.4 million network related gateway data points. These measurements were taken every 30 seconds from 16 randomly selected macOS agents (using our maximum data retention period of 30 days). Subsequently, the data points used were from agents when fully online with dualstack connectivity (i.e. can reach Internet based lighthouses), rather than just being locally_connected yet we only used the data from the client to the default gateway. Using Internet connected hosts makes any comparison fairer such that there’s the potential for ongoing traffic traversing connections and the gateways we’re measuring to. Additionally, it should be noted that PanSift does not normalize data and retains full fidelity metrics. This allows for fine grained analysis and even retrospective troubleshooting. After filtering data for agents with concurrent IPv4 and IPv6 Internet reachability, we were then left with 342,980 valid gateway related data points.

Under The Hood

The PanSift agent sends 3 ICMP echo_requests every 30 seconds using both IPv4 (ping) and IPv6 (ping6) while also explicitly setting the COS (Class of Service) to Best Effort . The options used with ping and ping6 (just to be explicit) are:

• -c 3 = to send a count of 3 requests and then stop
• -i 1 = explicitly wait 1 second between requests (default)
• -k BE = Best Effort normal class (default)
  
  Note: We explicitly use and mention the above defaults as placeholders and reminders to ourselves for potential further tweaking. We also used a parent process timeout of 5 seconds for ICMP to the gateway. IPv4 requests went first, then IPv6, and we recorded the average latency of the 3 requests as the resulting data point (using a float value in milliseconds).

With queries to our Influx backend (a Time Series Database), the data spanned macOS versions 10.11.x, 10.14.x, 10.15.x, 11.6.x, and 12.x with physical models ranging from iMAC to Mini but mostly Macbook Airs and MacBook Pros.

We then used Influx’s flux scripting language to do some simple comparisons on whether IPv4 or IPv6 was faster.

Note: To simplify, we also rounded the (float) values to (integer) whole numbers so we could quickly and more fairly compare “ties” between IPv4 and IPv6.

Results

Response Type	# IPv4 Faster	# IPv6 Faster	# Tied / Same	N (sample size)	Summary
WLAN + Wired	107,881	134,401	100,698	342,980	IPv6 wins!
Percentages	31.45%	39.18%	29.35%	100%	IPv6 won !

Table 1.0 - IPv4 vs. IPv6 Total Latency Results
Image 1.0 - IPv4 Total Latency Results

Image 2.0 - IPv6 Total Latency Results

Response Type	# IPv4 Faster	# IPv6 Faster	# Tied / Same	N (sample size)	Summary
WLAN (Wi-Fi)	97,638	127,335	76,731	301,704	IPv6 wins!
Percentages	32.36%	42.20%	25.43%	100%	IPv6 won !

Table 2.0 - IPv4 vs. IPv6 Latency By WLAN

Response Type	# IPv4 Faster	# IPv6 Faster	# Tied / Same	N (sample size)	Summary
Not WLAN (Wired*)	10,243	7066	23,967	41,276	Tied / Same
Percentages	24.81%	17.11%	58.06%	100%	Tied / Same

Table 3.0 - IPv4 vs. IPv6 Latency By Wired*
* We require a wider data set as most agent data was from Wi-Fi / WLAN connected hosts. The types of wired hardware recorded were primarily, "USB LAN" and "USB-C LAN" types, some of which may have been synching at lower than gigabit (1000 Mbps) speeds. As such, we need to also start to grab the media output, however both IPv4 and IPv6 are traversing the same media!

IPv4 and IPv6; Relevant Differences?

What is IPv6 connectivity anyway?

We’ve previously done a primer on IPv6 connectivity and troubleshooting here, so let’s take a brief look at some of the fundamental differences between IPv4 and IPv6 in relation to our data.

ICMPv4 and v6

The ICMPv6 format is described in RFC4443 whereas ICMP is described in RFC792, yet apart from the difference in Extension Headers (which are not relevant here), one wonders about why IPv6 might be faster than IPv4 (perhaps related to the gateway stack and utilization?).

Next Steps?

We now need to expand the above analysis to see if:

• there’s any correlation with the MAC addresses (and thus the vendor OUIs) which show what type/make of device is acting as the gateway?
• perhaps this is related to certain client device types, operating system versions, or patch levels?
• a larger sample set of dualstack devices will result in the same outcomes?
• are there ARP and NDP considerations (based upon traffic or lack thereof)?
• could the order of testing, serial versus parallel (or random) play an effect?
  
  Note: Based upon feedback to this article on the potential for sleeping Wi-Fi radios and differences with ARP and NDP, we are going to randomize which of the default gateway ICMP tests (in the agent) go first. Once a critical mass of the install base have the new agent scripts we will re-run the above experiment.