Results for reduced performance at lower data rates on 802.11ax in Linux
Results for reduced performance at lower data rates on 802.11ax in Linux
The problem with `iw wlp3s0 link` indicates the incoming signal tops out near 1.2-1.4gbps, while outgoing reaches about 1.7-1.9gbps. Iperf3 shows roughly 400mbps from desktop to router and 600mbps from router to desktop. It's unusual for the slower link to be 50% faster than the faster one. This pattern repeats on wired connections at 1gbps, and it continues even when TX drops to around 1.2gbps. For reference, a laptop on the same network with a Realtek 8812au chipset (802.11ac) achieves a TX of 867mbps, yielding similar ~400mbps (around 390mbps) in the same direction. What adjustments should be made? I expect at least 800mbps from desktop to router using this link. Hardware/software: Your NIC uses an Intel AX200 chipset, iwlwifi driver with firmware 63.c04f3485.0 cc-a0-63, Debian Bookworm with kernel 5.14.9 (5.14.0-2 from the repo). The router is an ASUS RT-AX3000 (same as RT-AX56U, though the model name might differ at Best Buy). I don’t know why this happens. Raw data/results: The iperf3 output shows bitrates and transfer speeds across various time intervals, with consistent performance around 400mbps in both directions.
Are you connecting your device to a 1 gigabit connection? You might need a modem with dual 2.5Gbps Ethernet ports. Your internet performance is limited by the Ethernet speed of your modem. Speeds you see are comparable to Wi-Fi 5, but it seems your Wi-Fi 6 performance is lagging behind due to the Ethernet bottleneck. If you have a machine with 2.5Gbps or 10Gbps ports, it would work better over Wi-Fi 6. In reality, your actual throughput is about half of the bandwidth, resulting in wireless speeds around 800-900 Mbps.
The speed measurement was just a sample. Most of the network isn't wired except for the modem upstream from the router. The Ethernet performance could have been misleading—perhaps I should recheck it later with a longer cable. The tests used the network idle and one-way at a time. Since the iperf3 server ran on the router, no extra hops were added. If I'm not wrong, the full wireless capacity should have been accessible. Edit: Frustrated. I should have run this sooner. Connected the laptop via Ethernet, tested laptop→desktop→laptop (still wireless). Maximum Ethernet speed (940mbps) was achieved between them. That rules out the desktop. It looks like the router is handling traffic internally might be the problem. I’d still like to fix that if possible, but it seems unlikely. Edited October 20, 2021 by ApeironTsuka More testing
I didn't mention any speed reduction in wired connections. What I stated was that wireless performance is usually around half the bandwidth capacity. Wireless AC generally caps at about 850mbps, even though theory suggests higher limits. Some bands reach 1700mbps but are constrained by the Ethernet port speed. I haven't verified this, but conceptually you could transfer wireless AX between two wireless AX devices. Your performance would be halved by the highest data rate allowed. I use a mesh network with wireless AC. My wired links run Cat6 with gigabit internet, placing me in the 800-850mbps range. My throughput typically falls between 450mbps and 500mbps. Without wireless backhaul, you'd see around 380mbps to 500mbps. If your connection offers 1750mbps, you should achieve roughly half of that—about 875mbps—in an iperf test. A gigabit link supports up to 940mbps.
You asked about changes made this time and clarified your concern about testing. The method differs because you’re now running the test locally, so results reflect actual device-to-device speeds rather than server-based measurements.
I was operating the iperf3 server directly on the router. There were no network connections nearby that could even reach AX, and I mistakenly believed the router wouldn’t cause any link saturation in either direction. The test focused on the speed between my desktop and the router, since I had a USB 3 drive connected as a NAS. Based on my earlier assumption, I didn’t consider checking the router’s performance.