Linux version 3.10.108 on stock
The link in the top post does not contain anything close to 450 Mbps with sqm on an tp-link AP/router...
Pretty sure that is too old:
https://www.usenix.org/system/files/conference/atc17/atc17-hoiland-jorgensen.pdf
Our modifications have been accepted into the main- line Linux kernel, different parts going into kernel releases 4.8 through 4.11, and is included in the LEDE open source router firmware from release 17.01. The implementation is available online, as well as details about our test environment and the full evaluation dataset.3
It's on a MT7621 aswell
Its been some time since. You can get same nowadays via soft offload. Patting self on the back
And bridger helps to get WAN port make friends with DSA LAN switch (50->10% CPU as measured on RT-AX54), still the separated WAN is disadvantaged in dumb config.
I wasn't questioning that the WiFi was not comparable, just the claim it would achieve 400 Mbps with SQM....
Gotcha, My bad...
Does having my raspberry Pi 4 running SQM on wan side make a difference in the configurations/load if I do sqm on the AP?
Your raspberry does SQM so well that you cannot improve over it using additional device on path.
I believe it needs to be done on the WLAN side?
You need to determine what held back TXOP for 100ms, like neigbor with huge antenna beaming wifi for kilometer away across your roof or so. i.e do wifi "site survey" mapping signal strength around home.
I am the only 5ghz network around, so....
1 Like
On the risk of sounding like a broken record, your best bet for WiFi is to:
a) use an AP that implements all of TXQS, AIRTIME_FAIRNESS, and AQL (likely your AX23 under OpenWrt will do)
b) for internet traffic only, set the wan sqm instance shaper rate below what your AP delivers worst case over the air.
1 Like
My unfounded conspiracy theory is that OEM disables (1) if you refuse to use their mesh app. Test would be to set TOS in VO or VI class and test latency again.
Different question, what are you using as computer to run your online tests via wifi?
1 Like
It's a vivobook laptop with a Wifi 6 Intel AX201
2 Likes
USe the mesh app for what? I got the tether app installed and working.
There is no compelling reason for such a high latency.
I did once get 3ms uptick on wifi.
But that only happened once.
+30 +150 in original post came from?
The "average" figure over wifi.
The 3ms was a one off
So which test are you referring to that reported 3ms uptick?
And keep in mind, that by default fq_codel/cake aim for 5ms target (WiFi codel even higher, I believe 20ms by default) so under saturating load with enough parallel flows the expectation is to see these 5 or even 20ms delay as latency increase under load...
1 Like