Don’t have screen shots to provide but for waveform all I can say is that my download active latency would spike crazy like up to a 100ms. Upload active latency numbers would spike but not as bad as download. On friendlywrt fw I never had an issue using 1400mb cap on sqm cake with the performance mod. For some reason when I used to rc openwrt versions and stable openwrt official release version I cannot do sqm that high even with a performance mod with packet steering as well. If I cap to 900mb download SQM works well.
Are you using the Realtek RSS drivers?
Yes the kmod-r8125-rss
Other than try to increase the amount of queues with Ethtool, the next step would be to check the CPU load during a speedtest with htop.
How do you increase the queues? I know about ethtool -k, c, a, and g
Nvm I’m assuming you’re talking about ethtool -G and increasing buffer sizes. Mine is set to default at 512 both ways for eth0 which is lan port for my pc. Both eth1 and 2 are set pre default to the max 1024.
What’s your output for the tweak? I’m struggling with sqm at 2Gb too
I just used @StarWhiz's guide and used the first option: https://github.com/StarWhiz/NanoPi-R6S-CPU-Optimization-for-Gigabit-SQM/tree/main
The output is all of the four faster cores being utilized, as visible in HTOP and also the Luci statistics graphs.
The Download bloat still a touch high (relatively), but can live with that. Just using Piece of Cake.
See that’s so weird… I used the first option too but my download active was going crazy. Like up to 100ms plus unlike yours that’s actually relatively stable at 15ms download active. Mind you that’s even just trying to cap at lower than 2GB like 1400
I was referring to ethtool -L, i have no idea if there is any ability to increase the amount of queues since i don't have a R6S or anything else that uses the Realtek driver. But it would be worth a shot to see if you can split the load out onto more cores.
Sounds like maybe need to mess with the SQM settings? Mine is just set at 1.8GB up/down. What is it like with SQM turned off?
I just got my nanopi R6S, installed OpenWrt 24.10.0 r28427-6df0e3d02a / LuCI openwrt-24.10 branch 25.014.55016~7046a1c but I have problem with the ports: If I assign eth2 to WAN interface (eth2 is preconfigured in WAN) it correspondents to the 2nd physical port whuch is LAN1. So I have to put the WAN ethernet cable to LAN1 port instead of the WAN port. Any solution?
eth interfaces are hardcoded as follows below. WAN should be assigned by default to eth1.
You might have a misconfigured network. Would you please share your /etc/config/networkcontents?
Anyway, these configurations below are the ones you should care to have the correct mapping:
(...)
config device
option name 'br-lan'
option type 'bridge'
list ports 'eth0'
list ports 'eth2'
(...)
config interface 'lan'
option device 'br-lan'
(...)
config interface 'wan'
option device 'eth1'
(...)
config interface 'wan6'
option device 'eth1'
(...)
config interface 'loopback' option device 'lo' option proto 'static' option ipaddr '127.0.0.1' option netmask '255.0.0.0' config globals 'globals' option ula_prefix 'fd00:ab:cd::/48' option packet_steering '1' config device option name 'eth2.835' option type '8021q' option ifname 'eth2' option vid '835' config interface 'wan' option device 'eth2.835' option proto 'pppoe' option username 'xxxxx@xxxxx.xx' option password 'xxxxx' option ipv6 'auto' config device option name 'br-lan' option type 'bridge' list ports 'eth1' list ports 'eth0' config device option name 'eth1' config device option name 'eth0' config interface 'lan' option device 'br-lan' option proto 'static' option ipaddr '192.168.1.1' option netmask '255.255.255.0' option ip6assign '60'
These were my settings. I configured it the way you describe. I thought WAN should be eth2. Thank you!
Ehh I’ve already have. I can set SQM to 900mb and I get A grade and no latency increase but anything beyond that is just horrid. Maybe hardware on my R6S is wonky? Idk
Does anyone when RSS support will be merged into openwrt for driver r8125-rss? Currently cannot customize RSS settings for 2.5G ports on the R6S.
For your own safety please edit and remove your personal data (e-mail, passwords, etc) from the post above.
Your configuration is non-standard, and your WAN interface is clearly referring to eth2 which is tne LAN1physical port.
I would recommend reseting to factory defatuls and starting from scracth.
Thank you, I had removed sensitive data, but because formatting in my post was not good I deleted the code and pasted again and then forgot to remove password again! I changed wan to eth1 and think it is ok.
Now it is:
config interface 'loopback'
option device 'lo'
option proto 'static'
option ipaddr '127.0.0.1'
option netmask '255.0.0.0'
config globals 'globals'
option ula_prefix 'fd00:ab:cd::/48'
option packet_steering '1'
config device
option name 'eth1.835'
option type '8021q'
option ifname 'eth1'
option vid '835'
config interface 'wan'
option device 'eth1.835'
option proto 'pppoe'
option username 'xxx@xxx.xx'
option password 'xxx'
option ipv6 'auto'
config device
option name 'br-lan'
option type 'bridge'
list ports 'eth0'
list ports 'eth2'
config device
option name 'eth1'
config device
option name 'eth0'
config interface 'lan'
option device 'br-lan'
option proto 'static'
option ipaddr '192.168.1.1'
option netmask '255.255.255.0'
option ip6assign '60'
1 Like
@Rome2 can you show the output of -
grep . /sys/class/net/eth?/queues/?x-0/?ps_cpus
cat /sys/devices/system/cpu/cpufreq/policy*/scaling_governor
cat /proc/interrupts
and also with the -rss are you using irqbalance?
hey just wanna ask your final settings -- its the optimization from starwhiz guide + packet steering? or is it without packet steering?