I have a strange behaviour I’m trying to get rid of for some weeks, but cannot get over it.
I have a Turris Omnia with OS 6.1. ISP is a Fiber network with 100MBit downstream an 10MBit upstream.
When Turris starts up everything is fine. Then over some hours the downstream get slower and slower and the latency increases. Tried with SQM (bufferbloat) but that did not help.
Measured speed over time (with my homeassistant):
Now the real strange thing is, that only the LAN connected computers suffering from this. Tries speedtest-cli from the turris itself (via SSH) show full speed.
When I reboot the Turris everything is fine again for some hours.
Netdata does not show anything special. Dont forget the weird thing, that speedtest-cli gets the full speed from the Turris itself.
I think must be LAN related, but i cant figure out what is responsible
Found out, that Turris itself degraded like the LAN too, if SQM is enabled. Disabling it immetiately releases full speed, but only for the Turris, not for LAN.
Following your hint did exactly that (tc): Noticed many dropped and overtimed pakets.
While playing around how to fix this, I enabled “Software flow offloading” in the Firewall config.
This seems to fix the issue, although I cannot say why.
Currenly working without SQM as the average usage is below 10% of the overall available bandwith.
Any more ideas on why this issue occurs in “normal” operation and why “Software flow offloading” fixes it?
Not impossible that your router is overloaded with running SQM at the desired rates, software flow offloading will reduce the CPU load a bit and hence might help…