Andrew

You need to re-enable packet steering. ``` config/network config globals 'globals' option ula_prefix .... option packet_steering '1' ```

sofirq and kworker per core, should be some load at softirq0 and softirq1 at same time can you make a screenshot from htop - f2 unhide kernel threads + enable...

I see balancing being anomalous. All softirq (lilac colour, nftables and qdisc) happens on one CPU while hardware interrupts (red) look ?balanced? Normal is like this:  Certainly hitting 100%...

@nbd168 ?

At least try to revert to shell script, my pic is from 23.05.3 https://github.com/openwrt/openwrt/tree/openwrt-23.05/package/network/config/netifd/files

What do you get from `ethtool -l eth0` ? Try making queues of each of types match number of cores and restart irqbalance.

Question is if it is a multiqueue device: - check `/proc/interrupts` for more than 2 interrupts for eth0 - `ls /sys/class/net/*/queues/rx-*/*_cpus` And in multicore case - if different devices get...

are you running irqbalance to make interrupts go to differen cores?

Try running irqbalance, it should be able to direct different irqs to different CPUs. ??s_cpus being 0 means queue follows IRQ.

It is timeout (long interrupt serviceing) where normal PC would have PCIe ethernet card connected, not the wire side.