There absolutely are, yes. In fact, there are a lot of patches that @pesa1234 is continuing to tweak and tune based on the most recent wireless test results we’ve been talking through.
At the moment, going back to even mainline snapshot builds will feel like a step backwards in terms of wireless latency.
Pesa’s builds tagged mtk contain, by my count, around 47 additional mac80211 (Linux wireless subsystem) patches as compared to snapshot, many by mtk, and around 78 additional mt76 patches, again, many by mtk. So these builds are a lot more than a minor tweaked snapshot and contain a lot of work to integrate and tweak these patches for our kernel.
@pesa1234 Could you help me understand this better? If I'm set to 160mhz and channel 100, for example, would the 'Use specific channels' value be 100 to match the channel? Or is there more to it than that?
I compiled next-r2.4.mtk at the latest commit fcef05c and am using it. As compared to the last builds I tried, most recently the pre-compiled next-r2.4.mtk.test2, my IPTV box no longer becomes unresponsive (requiring a restart), and two printers which were disconnecting and reconnecting, sometimes with deauth messages, are no longer doing so. I had previously tried all manner of advanced wireless settings to no avail. Perhaps this patch (mac80211: increase beacon loss) has something to do with it, or maybe it's something else. In any case, I will keep testing.
While I am enjoying this thread, a key thing to note IS "slower clients suffer higher latencies.". Ideally, if codel is working properly, this will be capped to about two txops max, 11-20 ms on the download, tops, somewhat dependent on the aql setting. On wifi with a fifo rather than codel, latencies can climb past seconds. (Starlink, before implementing fq_codel a few months ago, measured out at 7 seconds)
One test I use is capping the maximum bandwidth at the wifi layer to mcs0 or mcs1 using the iw tool.
A good test is also, with any one of these devices, through a wall, or at least 30 feet.
Here are some public details about starlink's progress here. A key statement that Elon made somewhere, was that, yes, indeed, "lower latencies mean operations start sooner". I have often tried to explain slow start in this way, and how all of our benchmarks are broken when compared to real traffic. Most traffic completes in well under 3 seconds. As much as I like the crusader test you are using, someday perhaps we can just zoom in on slow start more...
The last build is providing great results with WED disabled and no SQM.
The score is still "A" but I see an improvement in download latencies, especiall with my OnePlus 8T (80mhz client) which had a higher latency last time.
Yes, the procedure to update is right. Also if you want to come back to gl-inet version, flash from openwrt webui.
Regarding the unneeded packages, without configure they don't start or if you want to be more sure you can disable from LuCi (system-startup-disable) or from command line example: /etc/init.d/wireguard disable
Appreciate the work and time that you dedicated to these builds @pesa1234
In my case, with standard configuration the results are almost same (stable A score) with irq and wed off, but, turning off irq and turning on rps i have many more results with (A+)
Im playing warzone on wifi now and have really stable results.