Unlikely, the run time cost of SQM is caused by the qdisc that are exercised, while most changes affect how SQM configures and starts these qdiscs... So not impossible, but rather unexpected. I would wonder more about the receive side scaling setting. What does:
for file in /sys/class/net/*
do
echo $file RX rps_cpus
cat $file"/queues/rx-0/rps_cpus"
echo $file TX xps_cpus
cat $file"/queues/tx-0/xps_cpus"
done
Nope, that is IMHO solid release engineering, if a bug can be safely worked-around, but progress on fixing it has stalled it seems better to get the release out instead of waiting longer for Godot... Yes, unfortunate for users that use that functionality, and hopefully a fix can be found. But the alternative of delaying the release even further seems far worse. This is of course just my subjective opinion, others might disagree.
Well, the alternative is everybody "skipping"/not getting this release and actually testing it under real life conditions on a varied set of devices and configurations. IMHO the final steps of release maturation require a decent deployment, which for an opensource project like OpenWrt more or less means cutting a release to recruit sufficient many "testers". (Ideally everything with a release just works perfectly, but bugs happen)
Could you post the output of tc -s qdisc for both versions by any chance and if you are at it also the output of SQM_DEBUG=1 SQM_VERBOSITY_MAX=8 /etc/init.d/sqm stop ; SQM_DEBUG=1 SQM_VERBOSITY_MAX=8 /etc/init.d/sqm start
That might help figuring out what is happening here.
Upgraded 5 Meraki MR-24's, keeping settings. One had been running 21.02.0-rc2 since June, the others were still on 19.07.07. Touched LUCI Network->Interfaces page on these latter 4 to trigger auto-migration of bridge and ifname configurations.
Everything working perfectly after a day. Many thanks devs!
I have a very bad situation with 21.02.0 on ASUS RT-AC57U. Something is wrong with the network connection. Delays and speeds are insanely bad. I barely can open any website, speedtest struggles to even start, and opkg update can't sync packages because of connection corruption / timeout.
Hardware is not overloaded or so, CPU usage is nearly zero and RAM has almost 100 MB free. No clues in the logs.
I've tried to tweak all the possible settings, hopped back and forth to 19.07.8 (with full reset) several times to be sure that this is firmware issue. And yeah, the result is: 19.07.8 works excellent, 21.02.0 is unusable.
Can only suggest that maybe the DSA works wrong on this hardware? No other ideas.
Edit: I found the issue. The only suspicious setting I was forced to use: MAC override.
Override is needed for auth to my ISP. I just tested through the other router and yes: it causes all this trouble. Without override the connection works perfect. But with this option set connection became unsuable.
So this is kinda a bug report now.
Mmmh, one way of looking at that. Then again OpenWrt aimed at 1-2 releases per year, and we are at least at 1.5 years since the last release already.... So 'when it is ready' is not OpenWrt's actual goal, it just happens to look like that....
And do we wait to long then all the other packages gets to old.
Like OpenVPN got updated to include with TLS1.3 and EC a long time ago but OpenWRT got stuck with the old version in 19.07.
Not to mention the actual kernel. 21.02 has the 5.4 kernel but at release time the world started using the 5.10. As far as I understand there where actual thoughts to upgrade to 5.10 before rc1 but that idea was scraped. But as fast as the branch was released i February then the master got upgraded.
To be honest I think 21.02 will live fast and die young. But with the new DSA it is a must release to get it working into the future…
Hm. Wireguard cannot be installed in my Archer C7, because "kmod-wireguard ist not available in any repository."
Does anybody know a solution? Or do i just have to wait because that kernel module package is not uploaded yet?
OK. That's strange...
Installation via SSH works and wireguard is up and works.
It is just the installation via luci web interface which fails. Here is a screenshot: