ok, I'm VERY happy we have consistent (if still excessive) latency, at the bandwidths being achieved at distance. Still getting it down to 20ms for 4 stations is my overall goal for good multiuser performance.
How does it "feel" compared to when you were at 100ms?
Do you mean "the_wifi_interface" to the SSID wifi5 network and " the_other_wifi_interface" to the SSID wifi4 network?
80mhz wifi5 r7800
ath10k client
... and if you are more comfortable two brick walls and furnature away, cake 60mbit.
Please note this is very much just a test, putting the shaper on specifically to see what happens with ack-filter, sharing on the rrul_be test.
pics good, also, after:
tc -s qdisc show
I used this card for 5 years until fairly recently, and was comfortable with it. I changed it for the ax210 to test it and I stayed with that one because it gave me more speed for the files transfer with the server (in windows). The ax gives me almost 100 MB/s on large files both upload and download on the r7800
I don't play on computers or consoles so I hardly notice the latency.
I gave this card to a colleague and claimed it yesterday... He is not very happy. I'll get it back in a few days...
thx, then for having sufficient curiosity about the latency aspect of it all!
Today I have already burned my free time and they are already claiming me. Tomorrow I will test the data you have passed me.
You are right. For me all this is a hobby that I love, but that I barely know.
it has been a hobby that I know all too well, and have been trying to pass on the love for, so I can retire, knowing, it's in good hands. http://the-edge.blogspot.com/2010/10/who-invented-embedded-linux-based.html
cya tomorrow!
I want to be sure i' doing well.
i put directly on
.../build_dir/target-arm_cortex-a15+neon-vfpv4_musl_eabi/linux-ipq806x_generic/linux-5.15.69
just after make menuconfig and just before make world.
I don't know why but I feel like I'm doing it wrong.
here is my branch.
Anyone want to try this?
After untaring the tarball I shared (here in the forums) following you can find the steps to follow:
make target/linux/{clean,prepare} V=s QUILT=1cd build_dir/target-*/linux-*/linux-*quilt push -aquilt import path/to/untared/patch/filename1 -P generic/filename1quilt push generic/filename1quilt import path/to/untared/patch/filename2 -P generic/filename2quilt push generic/filename2quilt refreshcd -make target/linux/update V=smake -j $(nproc) worldI reckon I'm not missing anything, not at my computer right now.
I add more information, it restarted the other day with 17 days of activity, I am using the same image as in the previous post, 5.10. For some reason this image causes some connection failures in the discord channels. Performing another stability test without touching anything.
Added your two commits to today’s master and 22.03 builds. Additionally dropped irqbalance.
I’ll do some work with 5.15 hopefully this week. Going to look through @qosmio repo - hopefully Ansuel’s krait findings make it in to master here shortly. 
thanks!
I will follow your 5.15 work.
I have been able to apply filename1.patch, but the second one i get a failed.
asvio@VM-MINT-PC01:~/R7800/R7800-qosmio-5.15-qsdk11/build_dir/target-arm_cortex-a15+neon-vfpv4_musl_eabi/linux-ipq806x_generic/linux-5.15.69$ quilt push generic/691-net-reduce-napi-poll-weight-to-8ms.patch
Applying patch patches/generic/691-net-reduce-napi-poll-weight-to-8ms.patch
patching file include/linux/netdevice.h
Hunk #1 succeeded at 2523 (offset 83 lines).
patching file drivers/net/wireless/mediatek/mt76/dma.c
Hunk #1 FAILED at 636.
1 out of 1 hunk FAILED -- rejects in file drivers/net/wireless/mediatek/mt76/dma.c
Patch patches/generic/691-net-reduce-napi-poll-weight-to-8ms.patch does not apply (enforce with -f)
Is it safe to do "-f" enforce?
I did and got a successfully target/linux/update
Yes, it should be fine. You don't have an mt76. However, to be a hundred per cent sure, just keep the following code in it:
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -2458,7 +2458,7 @@ static inline void *netdev_priv(const st
/* Default NAPI poll() weight
* Device drivers are strongly advised to not use bigger value
*/
-#define NAPI_POLL_WEIGHT 64
+#define NAPI_POLL_WEIGHT 8
/**
* netif_napi_add - initialize a NAPI context
However, as I said, the code failing is for mt76 devices, don't worry too much in your case.
I've just had another reboot after more than 48 hours (qosmio 5.15-qsdk11-new-krait-cc). Not using irqbalance nor packet steering. No new files in /sys/fs/pstore this time.
Now I've switched to performance governor to see if it will continue to reboot.
With 5.15 no more messages
kern.warn kernel: [4849297.621504] br-lan: received packet on wlan0 with own address as source address (addr:xx:xx:xx:xx:xx:xx, vlan:0)
For the other four R7800s that currently run the older @ACwifidude's master build (without irqbalance and packet steering) there are no reboots for more than 15 days. I think this is really good stability.
They all use the default ondemand optimized settings.
No more irqbalance?
I'm seeing those messages too, limited to 2 devices - a moto G7 (Lenovo) and a Lenovo tablet.
I think Lenovo may have something wrong in their stack.
No more irqbalance. Multiple users reported a little more stability (ondemand CPU gov for ipq806x devices still has a persistent bug). Additionally NSS builds just need CPU primarily for wifi. I didn’t see any performance degradation with removal with limited testing on my own devices.
Irqbalance makes sense from the vast majority of linux systems. It just might be as helpful as we want it to be yet for our devices when the hardware offloading is enabled. Let me know if you see performance drop.
u doin the napi = 8 patch?