Wireless Ethernet Dispatch

[hpx502766238]

When connected to a 5G wifi and conducting a test with a download speed of 800Mbps, the top command revealed a CPU utilization rate ranging from 1% to a high of 56%.
However, with the official miwifi firmware, the sirq usage was consistently below 5%, with the overall CPU usage never exceeding 10%.
I believe this is one of the reasons for the occasional instability in wireless throughput.
Mem: 143668K used, 89208K free, 1908K shrd, 56K buff, 38796K cached
CPU: 0% usr 11% sys 0% nic 44% idle 0% io 1% irq 42% sirq
Load average: 1.03 0.55 0.26 6/85 10344
PID PPID USER STAT VSZ %VSZ %CPU COMMAND
611 2 root RW 0 0% 7% [napi/mtk_eth-4]
10 2 root SW 0 0% 2% [ksoftirqd/0]
6940 2 root IW 0 0% 1% [kworker/u4:4-nf]
20 2 root IW 0 0% 1% [kworker/u4:1-nf]
6939 2 root IW 0 0% 1% [kworker/u4:3-nf]
610 2 root SW 0 0% 0% [napi/mtk_eth-3]
650 2 root IW 0 0% 0% [kworker/0:9-eve]
10344 10336 root R 1256 1% 0% top
7 2 root IW 0 0% 0% [kworker/u4:0-nf]

[hpx502766238]

It seems lack of WED(Wireless Ethernet Dispatch).

[hpx502766238]

I conducted a comprehensive wireless performance test in WWAN networking mode. Below are the test conditions:

1. I used a PC1 equipped with a Killer WiFi 6E AX1675i (211NGW) network card to create a hotspot, simulating public WiFi signals. The card provides an approximate throughput of 800M after excluding interference factors affecting card performance.
2. Iperf3 -s was run on PC1 as the test server with the IP address 192.168.137.1.
3. The router connected to the PC1 hotspot through its 5G apclii0 interface, negotiating a rate of 1.1G.
4. The router's physical LAN port was connected to the test PC2, forming an internal network with the subnet 192.168.31.x/24.
   Below are the test results for wireless throughput and CPU utilization under different firmware:
   1. Regular firmware without Fast-NAT Support (using openwrt-mediatek-mt7622-xiaomi_redmi-router-ax6s-22.03.3-squashfs-sysupgrade.bin)
   1.1 Router as iperf client, testing throughput and CPU utilization.

1.2 Router as iperf client, Reverse mode (server sending data), testing throughput and CPU utilization.

1.3 PC2 as iperf client, testing throughput and CPU utilization.

1.4 PC2 as iperf client, Reverse mode, testing throughput and CPU utilization.

2. Firmware with Fast-NAT Support (WHNAT) (using openwrt-mediatek-mt7622-xiaomi_redmi-router-ax6s-22.03.6-squashfs-sysupgrade.bin)
2.1 Router as iperf client, testing throughput and CPU utilization.

2.2 Router as iperf client, Reverse mode (server sending data), testing throughput and CPU utilization.

2.3 PC2 as iperf client, testing throughput and CPU utilization.

2.4 PC2 as iperf client, Reverse mode, testing throughput and CPU utilization.

Test conclusions:
a. Comparing 1.1 and 1.2, we can see that high throughput is achieved in both directions for wireless Layer 2 forwarding, but CPU utilization is high, especially for soft interrupts (sirq). This indicates that wed (wireless ethernet dispatch) is not enabled, and wireless and switch packets are forwarded through CPU software.
b. Comparing 1.2 and 1.4, we can see that there is approximately 100M loss in downstream throughput when using Openwrt's built-in Hardware NAT, while upstream throughput remains almost unchanged.
c. Comparing tests 1 and 2, we find that in the firmware with WHNAT enabled, CPU utilization is significantly reduced (to single digits) only when the router handles NAT upstream forwarding traffic. For other forwarding traffic, soft interrupt utilization remains high, suggesting WED may not cover all scenarios.
d. There is also a serious bug where downstream throughput seems to be locked below 200M according to tests 2.2 and 2.4, which needs to be fixed.
e. I have also been studying the hwnat code myself but have only concluded that WED is part of WHNAT in the driver code. I'm not very familiar with hardware and am just a network engineer who has played with 51 microcontrollers before. I hope these tests can help you and that you will maintain your passion for technology.