is your eth0 which is 1g lan using the gmac driver??
and i know the numbers i listed are decent but these were consistent spikes for me since i know what my numbers looked like pre-r6s usage. whole idea of sqm is to keep numbers consistent to ensure there is no sudden ping spikes and jitter spikes but with the r8125 it was not doing that
Based on the following I'm thinking it's 8125, but I don't really know how to check. According to Luci I don't use eth0 and instead use eth2 for lan.
root@OpenWrt:~# lspci
0003:30:00.0 PCI bridge: Rockchip Electronics Co., Ltd RK3588 (rev 01)
0003:31:00.0 Ethernet controller: Realtek Semiconductor Co., Ltd. RTL8125 2.5GbE Controller (rev 05)
0004:40:00.0 PCI bridge: Rockchip Electronics Co., Ltd RK3588 (rev 01)
0004:41:00.0 Ethernet controller: Realtek Semiconductor Co., Ltd. RTL8125 2.5GbE Controller (rev 05)
Edit: I just realized I ran the wrong opkg command. I only have kmod-r8125-rss installed.
check with ethtool -i eth0
mine uses a gmac driver which i believe is embedded driver
Looks to be r8125, which makes total sense now that I see r8169 is not installed. Here are the results of a waveform bufferbloat test if you're interested. Jitter stays under 2ms.
root@OpenWrt:~# ethtool -i eth0
Cannot get driver information: Resource busy
root@OpenWrt:~# ethtool -i eth1
driver: r8125
version: 9.014.01-NAPI-RSS
firmware-version:
expansion-rom-version:
bus-info: 0003:31:00.0
supports-statistics: yes
supports-test: no
supports-eeprom-access: no
supports-register-dump: yes
supports-priv-flags: no
root@OpenWrt:~# ethtool -i eth2
driver: r8125
version: 9.014.01-NAPI-RSS
firmware-version:
expansion-rom-version:
bus-info: 0004:41:00.0
supports-statistics: yes
supports-test: no
supports-eeprom-access: no
supports-register-dump: yes
supports-priv-flags: no
When I am running r8169 (installed kmod-r8169 and removed kmod-r8125-rss, rebooted, and verified via ethtool -i) I get these waveform bufferbloat results. Which are pretty darn close.
looks like it cant get the stats for eth0 but im assuming its using an embedded driver and your tests are phenomenal im assuming you are on fiber internet??
Yep, Verizon Fios (1g symmetrical). Ok, I'm going to stop annoying everyone now 
. Sorry I couldn't provide any suggestions to help, @Rome2, but I hope r8169 continues to get great results.
its alright man thanks for sharing your experience brother
anybody else having this issue? cant seem to adjust values. i just set up kmod-r8125-rss to give it another shot
ethtool -l eth2
Channel parameters for eth2:
Pre-set maximums:
RX: 4
TX: 2
Other: n/a
Combined: n/a
Current hardware settings:
RX: 4
TX: 1
Other: n/a
Combined: n/a
root@OpenWrt:~# ethtool -L eth2 TX 2
ethtool (-L): unknown parameter 'TX'
root@OpenWrt:~# ethtool -L eth2 tx 2
netlink error: Not supported
maybe im having bad performance with r8125 since its not in the linux kernel. meaning it taints the kernel as it is an out of tree module?
This patch is required.
A pull request has been submitted, but it has not yet been merged.
so we cant quite make the changes yet till it is added into the kernel?
are you using irqbalance and packet steering?
In my case, I build it myself after applying this patch. The TX queue is 2.
# ethtool -l eth1
Channel parameters for eth1:
Pre-set maximums:
RX: 4
TX: 2
Other: n/a
Combined: n/a
Current hardware settings:
RX: 4
TX: 2
Other: n/a
Combined: n/a
Reading through recent posts, I've noted discussion of inconsistent gaming SQM performance with 24.10rc5 and earlier. I did some SQM tests on Waveform before and after this commit, which is in 24.10rc6 and later.
Compared to 24.10rc5, with 24.10rc6 SQM throughput not only increased by ~40%, the latency statistical distribution was also much tighter, with a lower median and no spikes. It was a huge improvement on my R5C. Yes, this is the R6S thread, but that commit made such a big difference for me, it may be worth your time to try an upgrade to 24.10rc6 or later.
I agree, i'm using a R6S on rc7 currently, I've used 8125-rss and 8169 drivers today to test, and i'm not noticing much if any difference. The only noticeable thing is that with the 8125 drivers the eth's are spread over all cores, with 8169 the are only on 2. I have spread queues over all cpu's. and i've tried with and without irqbalance, the 8125 drivers IMHO work best with this enabled, and that's purely from throughput perspective watching htop and the cpu load's.
i will add ive got this in my startup -
find /sys/devices/system/cpu/cpufreq -name scaling_governor -exec awk '{print "performance" > "{}" }' {} \;
running the cpus on performance.
1 Like
Iām using packet steering as of right now
How would I go about doing it myself?
See gaming performance involving SQM was horrible for me on default friendlywrt firmware and on the rc images. It was only because of the kmod-r8125 driver I noticed this behavior on my system. Good to know that even with the r8169 Iām still pushing 2 gig speed.
1 Like