netisr saturates one CPU core, regardless of traffic. DOCSIS not PPP.

Started by CloudHoppingFlowerChild, November 27, 2022, 08:46:50 PM

Previous topic - Next topic
This issue has followed me from 21.1 on hardware to 22.7 on ESXi 7. After approximately 2 minutes of up time, one CPU core will be maxed out. Systemâ–ºDiagnosticsâ–ºActivity shows netisr is the culprit. Googling suggests this is usually due to high traffic yet this exists regardless of traffic level and it doesn't seem to limit throughput in any practical way; though my downlink is only ~275mbps. I have tried stopping all the services I could and that didn't cause any change. Restarting Packet Filter, System routing, System turntables had no noticeable impact (though its possible they restarted so quickly I wouldn't have seen an impact). It's almost a cosmetic problem aside from extra power consumption. Any thoughts or suggestions on where the problem might be? I'm thinking bad configuration on my part, somewhere, has lead to something going in circles but not sure where.

I had the same issue with PPPOE connection, I fixed it with 3 tunables settings (might not work with DOCSIS but try it) :


net.isr.bindthreads = 1
net.isr.maxthreads = -1
net.isr.dispatch = deferred


Please read this:

https://forum.opnsense.org/index.php?topic=24409.0

This net.isr.dispatch = deferred is AFAIK only needed for PPP/PPPoE, in all other cases it is not necessary to set this.


In theory there is no difference between theory and practice. In practice there is.