Johnny Vogels

Not sure if it's going to work though in Fenics without dual basis.

See also https://arxiv.org/pdf/1807.01416.pdf (Hexahedral div elements) and https://www.doi.org/10.1137/15M1013705 (AC div elements) and https://arxiv.org/pdf/1809.02192.pdf (direct serendipity)

Do you really want to have this merged now?

Packet steering was always disabled. Packet steering appears to do something, but not for uploads. And also downloads is still slower under circumstances (iptv). From top, it appears all irqs...

Bisecting in progress to between last commit april 13 c2f52e42b1e9f736edd6e4545ebe3435db6605ee good last commit april 14 cab2e1de0d94ab35b5d032194fc5b4ab2e0d609a fail (both on kernel 6.1) htop (with packet steering to all cores and software...

Following https://natsys-lab.blogspot.com/2012/09/linux-scaling-softirq-among-many-cpu.html improves things a little bit further, but cpu0 is still hit harder.

The problem is indeed this commit a205a5734eda4604a19b6cb0f65909ef69730699 @nbd168 .

Maybe packet steering to all cpus works... (packet steering = 2) Checks continuing. For now it seems that Luci having packet_steering = disabled appears the same as enabled. This removes...

Still degraded performance on the latest. Going to do more analyses, bisections and back-porting. But the packet steering disabled = enabled issue is definitely something that should not be happening.

Ok, so issueing `/usr/libexec/network/packet-steering.uc 0 -d` (This can not be issued from luci) resolves the situation, while it lasts. Somehow `/usr/libexec/network/packet-steering.uc 2 -d` does not have the same effect, even...