The results: https://i.imgur.com/gsQAHYP.png
I think this helped for 2-4% or so, thanks!
I have a question regarding "Maximum transmit power" under wireless.
When setting it to 30 dBm, sometimes "Current power" reads "26 dBm", and different values other times.
The description underneath reads: "Specifies the maximum transmit power the wireless radio may use. Depending on regulatory requirements and wireless usage, the actual transmit power may be reduced by the driver."
I wonder what is affecting the different power output? Just an example, I am trying channel 36, 160 MHz, and no one else is using that 160 in my area. I read that people are using 30 dBm with above settings.
Is there anyway to force the output to always use the "max value" allowed inside the field? Most search results talk about people trying to tweak the max power beyond the displayed values. However in my case, I can choose 30 dBm as max, but the "current power" is somehow lower.
The random "current power" does not happen in 160 MHz, but also in 80 MHz and even in 2.4 GHz.
Thanks.
I'll be curious to see the results after you get your fiber installed! I notice your loaded latency dropped fairly significantly, so that should make things snappier on your network overall. That was a noticeable result on my own system as well. Changing your processor governor to performance would be another test, but I'd wait until you get the faster speeds to see if it's necessary or not.
Any idea how to set the Tx-Power to more than 17dBm? After changing the country settings to german I always get 17dBm no matter what kind of Tx-Power I choose …
5ghz or 2.4ghz?
What channel are you using?
5ghz Channel 36 - should allow up to 23dBm in Germany…
Hmm strange. Does it exhibit similar behavior on other channels?
Jup other channels as well …
Will try to use other country codes ab see what happens.
Just notice your tcl values are different from ax3600 thread:
https://forum.openwrt.org/t/adding-openwrt-support-for-xiaomi-ax3600-part-1/55049/4804
root@D1:~# cat /proc/interrupts | grep reo2host-destination-ring
50: 3118 0 0 0 GIC-0 320 Edge reo2host-destination-ring4
51: 6461 0 0 0 GIC-0 271 Edge reo2host-destination-ring3
52: 17783 0 0 0 GIC-0 268 Edge reo2host-destination-ring2
53: 3661 0 0 0 GIC-0 267 Edge reo2host-destination-ring1
root@D1:~# cat /proc/interrupts | grep wbm2host-tx-completions-ring
73: 55608 0 0 0 GIC-0 191 Edge wbm2host-tx-completions-ring3
74: 66363 0 0 0 GIC-0 190 Edge wbm2host-tx-completions-ring2
75: 403793 0 0 0 GIC-0 189 Edge wbm2host-tx-completions-ring1
I don't know, but maybe the same ring # should run in same core?
1 Like
I am using HK, 5GHz only:
Channel AX, auto, 160MHz
(cannot verify 160MHz is working or not, no device capable)
Power default
Luci reporting 24dBm, Channel 36
1 Like
Good point. I’ll shuffle them to the same core when I get home. I’m not an expert, but I could see how beginning in one core and ending in another could incur a slight latency penalty. Thanks!
EDIT - It's funny... I realize I had it set differently at first, then without even thinking about it I reset it to the way I posted because after a reset about a week ago, in my mind I was staggering them evenly. Initial way I had it (and the way the Xiaomi folks use it) is listed below. @borgqueenx - I'd update yours to match this as well. Sorry for the confusion!
#assign 4 rx interrupts to each cores
echo 8 > /proc/irq/50/smp_affinity
echo 4 > /proc/irq/51/smp_affinity
echo 2 > /proc/irq/52/smp_affinity
echo 1 > /proc/irq/53/smp_affinity
#assign 3 tcl completions to 3 CPUs
echo 4 > /proc/irq/73/smp_affinity
echo 2 > /proc/irq/74/smp_affinity
echo 1 > /proc/irq/75/smp_affinity
3 Likes
thanks for the update.
Results:
1 Like
With "US" it's working like it should. As soon as I use "DE" as country code the Tx-Power is magically set to 17dBm ... There is obviously something wrong.
if you search on the ax3600 and other ip807x there is an outstanding ask from @robimarko for an updated and external regdb.bin. The issue is that the firmware doesn't have an up to date regdb is with kvaloe / qcom / qca to fix it.
1 Like
I've seen this patch to irqbalance, this will affect ipq807? https://github.com/openwrt/packages/pull/20501
I tested it also on my own dl-wrx36, but did not see any effect.
This is also an aarch64 device, so the upstream patch might have helped.
On rt3200 the patch clearly helped irqbalance to recognize new irqs that it could try to balance.
I wonder if part of the reason are the additional chips that take care of part on the network functionality.
1 Like
I think we must open an issue at https://github.com/Irqbalance/irqbalance/
I'm out on mobile. If nobody does it before I will try to open the issue this week.
It's useful to enable htop option to show the cpu freq. (see the highlighted line bellow). Also take a look at the other options.
The cpu will run at different frequency steps to save power, it can go from 1Ghz up to 2.2Ghz.
When you have a 85% core load, the only way to know if the core is almost maxed out or half free is to display the core frequency.
You only would see a speed increase with the new IRQ affinity if your core 0 was maxed out before (100% @ 2.2Ghz). The new IRQ distribution releases some load from core 0 to the other cores.
It doesnt help, irqbalance still sees all IRQ-s on ipq807x as class 0 and therefore doesn't do anything with them
2 Likes
Here are the test results, what does this tell you?
EDIT: That the advertised 2.2ghz is not reached, meaning i could probably get quite a bit more speed using wireguard, because the cpu decided it doesn't need such a clock speed?
1 Like