AirPort Extreme AC (6th gen) actually supports 1 Gigabit wired throughput speed on IPv4, but not IPv6!

Question

Level 1

8 points

AirPort Extreme AC (6th gen) actually supports 1 Gigabit wired throughput speed on IPv4, but not IPv6!

So I finally upgraded to a gigabit internet connection. I had previously rewired my house with quality Cat 6 cables to I was curious to know how good a job I did. My internet setup is a direct fiber optic line from the node to the ISP gateway/modem (in passthrough mode) to the AirPort Extreme serving as the router/DHCP server and WiFi access point. All copper connections are via at least Gigabit Ethernet ports and Cat 6 terminations.

In all the forums I've come across, everyone references the SmallNetBuider review claiming a max throughput between 300 - 400 Mbps in router mode or 500 - 600 Mbps in bridge mode.

Attached are my Speedtest results using 4 different servers: 2 IPv4 and 2 IPv6. The bottom 4 results (10:xx am) are a direct wired passthrough connection from the modem to my Mac. The top 4 results (11:00 am) are a wired passthrough connection via the AirPort Extreme to my Mac.

My ISP advertises a speed of 940 Mbps. As expected the bottom 4 results are all consistently around the max speed. The top 4 however diverge based on whether the testing server established an IPv4 or IPv6 connection. The 2 IPv4 connections result in download speeds of 918 and 947 Mbps. The 2 IPv6 connections result in download speeds of 647 and 644 Mbps.

So clearly the AEAC actually supports gigabit speeds. My fastest ever recorded speed is actually 947 Mbps, via the AirPort Extreme! Firstly, to me this brings into question the veracity of SmallNetBuilder testing and reviews. Second, and more importantly for me, what is the Airport doing with IPv6 connections that is causing the speed to reduce to only 2/3rds of the IPv4 connection speed? Can I change any settings to actually get full speeds via IPv6?

My IPv6 setup in Airport Utility is Automatic in Native mode and IPv6 Connection Sharing enabled. IGMP Snooping is enabled and incoming IPv6 connections are blocked.

This was quite a surprise to me that a 10 year old router can support this kind of throughput. I'm really curious why others are not able to achieve this as well. But I'm happy to add this datapoint!

Posted on Sep 15, 2022 10:55 AM

Reply

Answer 1

Top-ranking reply

Tesserax

Level 10

152,103 points

Sep 16, 2022 9:45 PM in response to tennisproha

Ok, those results look good.

To simply testing with iPerf3, especially the WAN-to-LAN interface, they provide public iPerf3 servers for your use.

For example, one server located in Los Angeles, is: la.speedtest.clouvider.net

You would then test from a client on your local network to that server ... and this test can be performed in both directions.

To test IPv4:
- Server-to-Client: iperf3 -c la.speedtest.clouvider.net -p 5200-5209 -4 -R << Measures download speed from the server.
- Client-to-Server: iperf3 -c la.speedtest.clouvider.net -p 5200-5209 -4 << Measures the upload speed of the client.
To test IPv6:
- Server-to-Client: iperf3 -c la.speedtest.clouvider.net -p 5200-5209 -6 -R
- Client-to-Server: iperf3 -c la.speedtest.clouvider.net -p 5200-5209 -6

If you want more details in the test results, add: -V

FWIW, I use a pair of Raspberry Pi 4's for this purpose. They are easy to setup at various locations on the network to facilitate running these tests.

Ref:

iPerf3 - Man Page - mankier.com
Public iPerf3 Serverlist - GitHub

Reply

Answer 2

tennisproha Author

Level 1

8 points

Sep 18, 2022 7:38 PM in response to Tesserax

I don't quite understand how putting the AEAC in bridge mode to test the WAN-LAN performance is any different than the LAN-LAN test I did earlier? The Mac and ATV would still be connected to the LAN ports, not the WAN port...

So someone elsewhere gave me a likely explanation for what's going on here:

The AC AirPort Extreme got to line rate for IPv4 NAT by relying on hardware acceleration (TCP and NAT offload to hardware). The Broadcom ARM-based SOC it used was not capable of offloading IPv6 traffic processing to specialized network acceleration hardware elsewhere in the SOC, so that traffic burns more CPU cycles and doesn't reach the same max throughput.

This makes a lot of sense to me and is probably what's going on considering Apple's implementation of IPv6 was sort of an afterthought in the first place.

What this doesn't quite explain is why I'm getting max throughput on IPv6 when I run the LAN-LAN bandwidth test via iperf. But I think maybe because there's no WAN port involved, Apple routes the traffic differently...

Reply

Answer 3

tennisproha Author

Level 1

8 points

Sep 19, 2022 3:10 PM in response to tennisproha

Just wanted to add to my previous answer for posterity. It looks like I was on to something regarding why the IPv6 throughput is lower on WAN but not on LAN testing.

Someone else gave me this explanation:

Your local host and server wont be creating connection tracking (Stateful firewall entries) and checking each packet against it. This is usually what drives CPU usage up on high PPS (packets per second) as each packets header has to be checked. IPv6 headers are larger than IPv4 headers, thus needs more CPU+RAM.

I think that answers it. The lack of hardware acceleration for IPv6 on AEAC reduces throughput speeds. Otherwise, the AEAC is actually capable of 1 gigabit bandwidth on a wired connection. And on Wi-Fi, in optimal conditions, I clocked max throughput speeds over 600 Mbps.

Thanks to everyone for helping me test and understand the results.

Reply

Answer 4

Tesserax

Level 10

152,103 points

Sep 15, 2022 12:05 PM in response to tennisproha

I see that you used Ookla's Speedtest for your measurements. Have you done similar tests totally within your LAN network? That is have you tested both the bandwidth & throughput between your AirPort Extreme and your Mac using something like iPerf? How about your Ethernet settings? For example, are you using the default Ethernet frame size or jumbo frames?

AFAIK, Apple does not impose limits based on specific IP versions. If anything, IPv6 does not utilize NAT, so the base station's NAT should not come into play at all, as far as, a restriction to traffic.

Reply

Answer 5

Tesserax

Level 10

152,103 points

Sep 15, 2022 12:47 PM in response to tennisproha

Sorry, I had a few other questions before my last reply was closed to editing.

If I understood you correctly, you did the Ethernet wiring ... correct? If so, and just curious, did you use solid or stranded wires? You mentioned that you used "Cat 6 termination." Did that include both Cat 6 male & female connectors (they should have small separators to prevent cross-talk between wires) at the appropriate end-points? Finally, did you test these lines? If so, how?

Chances are that your Ethernet runs are fine, just wanted to rule anything out for why you are seeing the throughput discrepancy between IP standards.

Reply

Answer 6

tennisproha Author

Level 1

8 points

Sep 16, 2022 9:58 AM in response to Tesserax

Hey sorry for the delay. I haven't done any LAN testing, mostly because I didn't have gigabit internet when I installed the cabling so I didn't wanna go about figuring out how to do it at the time. I'll look into iPerf this weekend. The ethernet settings are all default at the standard frames.

Yes I did all the wiring and terminations myself, using professional grade Belden supplies. The cables are Cat 6 solid bare copper. The jacks are Cat 6+ and the connectors are Cat 6A, all have separators to prevent crosstalk. So the cable itself is the weakest point as far as rating goes but it's very good quality so the lines should support at least 1 Gb if not way more. I didn't test the integrity of the lines themselves because that equipment is very expensive, like Fluke. I just made sure to do the wiring and terminations really well following best practices, so I'm pretty confident it's solid.

It honestly has to be the AEAC that's limiting it because the bottom 4 speed test results, which were direct from the modem to the Mac, were also split between the same 2 IPv4 and 2 IPv6 servers. The 2 IPv4 connections had download speeds of 849 and 943 Mbps. The 2 IPv6 connections had download speeds of 933 and 929 Mbps. So removing the AE from the flow resulted in full throughput on IPv6.

I'm not even sure what IPv6 settings to try to change in Airport Utility since I don't know what it's doing. However I do know that AE's don't support IPv6 on a guest network because of how it's configured so maybe Apple engineers just did something wonky to enable IPv6 support in the first place. All the tested devices are on the primary network btw.

Also, as a side, on WiFi I tested these same servers and got similar discrepancies, although not as drastic. With the iPhone showing a data rate of 866 Mb/s on 802.11ac, the IPv4 servers resulted in speeds between 400 - 600 Mbps, while IPv6 servers resulted in speeds between 300 - 500 Mbps.

If there's anything you'd like me to test or have more questions, let me know!

Reply

Answer 7

Tesserax

Level 10

152,103 points

Sep 16, 2022 10:56 AM in response to tennisproha

Yes, unfortunately good testing equipment comes at a cost.

I'd be interested in your iPerf results, especially if you have multiple servers. Ideally, you will want to test both the LAN side, as well as the WAN side.

For the LAN-side, I suggest that you set up one server as the iPerf "server," and another as the "client." Not sure how familiar you are with iPerf, so if you need any further clarification on its use, just let me know. iPerf runs within the Terminal app, but there is a GUI version, called jperf, that can be used (it still uses iPerf to perform the tests.) You can then try different port combinations (on the base station) with your two servers to see if there are any significant differences with any of them. There shouldn't be. If there are, then the base station's built-in Ethernet switch if faulty ... or, at least, one of its ports is.

For the WAN-side, you will need to place one server upstream of the base station to test the WAN-to-LAN throughput. This won't be necessary if you configure the base station (temporarily) as a bridge as all of its ports would just be Ethernet ports and not WAN or LAN-specific.

Reply

Answer 8

tennisproha Author

Level 1

8 points

Sep 16, 2022 7:10 PM in response to Tesserax

Attached are the iperf3 results.

I don't have another Mac so I linked between my Mac and Apple TV 4K, both wired via ethernet to the AirPort Extreme. I tested as both IPv4 and IPv6 servers. Surprisingly the results were about similar. IPv6 was marginally slower than IPv4 but both always came in over 900 Mbps no matter what ports I tested from.

This was LAN-side. I tried testing WAN-side but I just couldn't get it to work. Used both Mac and ATV as the WAN server. The Mac was self-assigning an IP address but not on the ATV.

As a side, what's interesting is the weakest bottleneck seems to be at the ethernet ports on the AEAC itself (and probably ATV4K), since they are only gigabit rated. I'm sorta curious now to test directly between the ATV4K and Mac using like a Cat 6 coupler just to see was the max throughput on the cables is. But I don't know how I would go about assigning IP addresses in that case either.

Anyway, still stumped in terms of why IPv6 is so much slower on the ISP speedtests.

Reply

Answer 9

tennisproha Author

Level 1

8 points

Sep 18, 2022 11:00 AM in response to Tesserax

I'm curious, how is this different than testing with Speedtest.net servers?

Wouldn't the WAN-LAN setup need to be local (in the immediate proximity) to give the most accurate bandwidth results...

But I tried the public iperf3 servers. The LA servers gave really wonky results. The Dallas servers gave more realistic ones. Here are the averages:

IPv4 client to server: 913 Mbps

IPv4 server to client: 507 Mbps

IPv6 client to server: 643 Mbps

IPv6 server to client: 333 Mbps

So the upload speeds are consistent with the Speedtest.net results, but the download speeds are half of both IPv4 and IPv6. But neither match my LAN-LAN throughput results.

I'm not sure what to make of this, especially the download speeds.

Reply

Answer 10

Tesserax

Level 10

152,103 points

Sep 18, 2022 1:10 PM in response to tennisproha

tennisproha wrote:

I'm curious, how is this different than testing with Speedtest.net servers?

In theory, it is just another method to do the same thing; however, you have more control over the testing parameters using iPerf vs. Ookla's SpeedTest.

I mentioned the public iPerf servers so that you can perform those WAN-to-LAN tests without having to configure a test rig on the WAN-side of your base station. The alternate method, as I mentioned earlier would be to temporarily isolate the base station by disconnecting it from the WAN. Then reconfigure it as a bridge so that it performs as an Ethernet switch. Then connect your iPerf client & servers devices to the base station to test each ports' throughput one at a time.

Something else to try with the base station configured normally, is to try different IPv6 settings to see if they help at all. Some will completely block IPv6 traffic so don't be too surprised at your results.

Finally, to be fully honest, I would suggest that you just retire your base station. FWIW, I had a number of them, but since Apple decided to vacate the networking hardware business in early 2018, I migrated to Ubiquiti's UniFi ecosystem with only a single 802.11ac Time Capsule still running so I can help folks with Time Machine issues here at the ASC.

Reply