I have a gl.inet a1300 slate plus with ipq4018. i have an AC 5ghz 80mhz access point on a dfs channel with no overlaps with WMM enabled. system logs show this as the command used to start the access point.
i have a desktop with ax210 (wifi 6) and mini pc with wcn6856 (wifi 6e). Both clients are connected to the access point and are hosting an ftp server.
The problem is the max speed I can get for file transfers is 30MBps. I thought maybe it was just the ftp server/client, so I tested with iperf3 and got the same results. I thought maybe an updated build would fix it so I built a custom 6.1 image and set everything up the same way. I am still getting these wifi speeds and it's irritating.
I tried enabling software flow offloading and hadware flow offloading and this made no difference.
What could be causing such slow wifi speeds? I'm at a loss of where to look.
Please connect to your OpenWrt device using ssh and copy the output of the following commands and post it here using the "Preformatted text </> " button:
Remember to redact passwords, MAC addresses and any public IP addresses you may have:
yep, i added it
i removed references to my wireguard interface but there are still some. it's wg2. it is in the wan firewall and was disabled during testing. that is why the wifi station connection has a metric set. before I setup the wireguard interface at all I was having the same results.
yes, the access point is the 5ghz band of the gl.inet openwrt router.
both clients are hosting an ftp server. actually i have about 5 clients all running debian sid all running vsftpd as a server and no matter which computer tries to copy files to any other computer, the speeds range from 14MBps to 30MBps, typically around 20MBps. i've tried multiple different pcie and usb wifi adapters, so I believe I've ruled out the issue being the clients.
vsftpd just allows access to the entire root directory if authenticated. My test was using a single 14.1GB file.
i didnt build in htop but top shows ksoftirqd reached 9% while about 4 others sat around 1-3%, but cpu % remained at 0% and user % hit 7%. Not entirely sure how to read top. This current test is averaging 19.7MBps.
I thought that may be the case and that is fine. The name is just a name.
I spent hours yesterday trying to get someone else to config a lan port as a wan port and could not get it to work for them, so I was hesitant qualifying it.
doing another build will have to be another day. and I have luci statistics app that has been graphing cpu usage, and the first core hasn't gone over 40% while the other 3 haven't gone over 8%.
edit
i went ahead and sshd to build pc and added htop, but i cant flash until tonight.
I recorded htop while running 3 tests with iperf3. Here's the video https://www.youtube.com/watch?v=CB6LlWQokAY
The cpu doesn't seem to be the bottleneck, although it is getting pretty high. This tells me it's either wifi driver/firmware or network configuration related.
im compiling now with lto, -O3, and cortex-a7 optimization flags to maybe get the cpu usage down. I also enabled the proprietary functionality option, i don't know if this will be relevant with the CT firmware. if this doesn't work, then I'm going to try the regular ath10k firmware instead of the CT one (CT is used in stock openwrt)
edit: lto is giving me trouble, ill post back later my results with non-ct firmware
I built with ath10k non ct driver and firmware and I have no improvement on wifi speed (21.9MBps iperf3)... So I think that rules out the problem being driver/firmware. The CT firmware was from 2020, the regular one is from 2023 so that's a plus. It's probably a good idea to move stock openwrt from CT drivers and firmwares, they apparently don't receive updates for long periods of time.
The cpu % went up to about 87% this time, but averages around 74% most of the time.
Turning off software and hardware flow offloading lowered the cpu to max 77% and average 67%, with no difference in wifi speed.
I'm currently trying to compile with some cortex-a7 tuning options and enabling the fpu with hard float, but it's proven quite challenging. I doubt optimizing the cpu is going to work. there's something else going on.
All of my client PCs have nvme/sata ssd's with minimum 100MBps read/write speed. With my old x86 router, I could saturate the 2.5GBe line between 2 PCs copying a file through FTP. I know for certain disk speed is not the bottleneck.
I tested that setting and I'm getting average 25.9MBps but the CPU usage seem to split between the 2 cores, 1st core averages around 50% now. CPU is definitely not the bottleneck now that I've seen that. Is it best to keep that setting on?
I don't suspect the driver anymore. My old router used a WCN6856 with ath11k on like the first kernel the device was available and was able to achieve 95MBps with hostapd, albeit not stable (it was a debian install). And I changed driver and firmware to the latest ath10k instead of CT. I'm thinking it may be wpad/hostapd or the configuration being used for the device sent to it. That's the last thing in the chain I can think of being the issue, except network configurations I don't know about.