On 2018-06-25 13:17, Bengt Gördén wrote:
Den 2018-06-25 kl. 11:56, skrev Carlos E. R.:
Interrupts are tied to a certain core? I thought they were dynamically assigned to a non busy core.
Yes. Your right. I think I was a bit unclear. What I meant was that if you have the need to redirect one interrupt source to a certain core it can be done via smp-affinity. Normally you shouldn't need that but there are cases where it is of great importance. The two cases I referred to (router and DAW) is such cases. They are real time cases. Under heavy load in a recording (audio) situation you do not want disturbance in any way and as low latency as possible. In the case of router it's crucial if you have a production system and you don't have an out-of-band solution to access the router and the traffic is very high. We did this in three different projects about 11-13 years ago where we showed that 10Gbit/s forwarding in a Linux router is possible. There we redirected the ssh-port to CPU0 and the rest of the packets were handled by core CPU1-CPU7 so that the control plane were separated from data plane.
I see :-)
But to get back to the "watch cat /proc/interrupts". It's to see were the interrupts comes from.
Yes, it is a good idea. Long ago, my computer was running very slow one day (perhaps SuSE 7.x). The process "init" was very, very busy. Single core CPU. I don't remember how I found out, but the modem was sending a flurry of interrupts over the serial port. I power cycled the modem and problem solved. :-) It never happened again. -- Cheers / Saludos, Carlos E. R. (from 42.3 x86_64 "Malachite" at Telcontar)