Latency, error finishing read, and IRQ affinity

Thanks mozmck for all this info and the bash scripts! This solved the weird issues I was running with on this new N100 PC I have.

I decided to brush up on my bash skills and update the original script.  

Features

• Checks that script is being run as root/sudo.
• Attempts to ping the common Mesa card IP addresses (192.168.1.121 and 10.10.10.10) and if it finds a unique one it will set that as the NIC variable in the irq bash scripts. If it doesn't find one it lists the possible options for the user to choose.
  • Downside of this is that it is not as dynamic as mozmck's script if there are any Mesa card interfaces changes, but I figured people rarely change that anyways.
• Give instructions once the script is complete.

How to run

• Rename rt_setup.sh.txt to rt_setup.sh
• Make sure script is executable via command line:
  Code:

  chmod +x /path/to/rt_setup.sh
  Replace /path/to/rt_setup.sh to where the actual file is.
• Run script: 
  Code:

  sudo /path/to/rt_setup.sh
  Again replace to where the actual path of the script is

Hopefully this helps those that are not software savvy!

I just discovery tuna, realtime IRQ tuning, my first impression was great.
Just test it at buster, going to bookworm now.

packages.debian.org/bookworm/tuna

Ran into an issue with IRQ balance after updating to trixie

Nov 16 19:34:22 hazelnut irqbalance[597]: IRQ 9 affinity is now unmanaged
Nov 16 19:34:22 hazelnut irqbalance[597]: Cannot change IRQ 14 affinity: Read-only file system
Nov 16 19:34:22 hazelnut irqbalance[597]: IRQ 14 affinity is now unmanaged
Nov 16 19:34:42 hazelnut irqbalance[597]: IRQ 123: Override ban to true
Nov 16 19:34:42 hazelnut irqbalance[1572]: /etc/linuxcnc/lcnc_irqbalancepolicy.sh: line 17: /proc/irq/123/smp_affinity: Read-only file system
Nov 16 19:37:52 hazelnut irqbalance[597]: Cannot change IRQ 135 affinity: Read-only file system
Nov 16 19:37:52 hazelnut irqbalance[597]: IRQ 135 affinity is now unmanaged



Turns out there was a bug version of irqbalance that is fixed with

github.com/Irqbalance/irqbalance/issues/308#issue-2217868587

after adding

/etc/systemd/system/irqbalance.service.d/override.conf

➜  ~ sudo cat /etc/systemd/system/irqbalance.service.d/override.conf
[Service]
ProtectKernelTunables=no

It fixed the ethernet mesa issue that was causing latency spikes





 

I’ve just updated to a Ali 7i92 with a Gecko540 on a old i3 pc, Plenty of ddr3 here. Running rt_setup allows the steppers to run smooth. Without it the motors appeared to be missing some steps. I had some problems running the script to find its a download error, dos2unix fixed the file to run as above. 
Thanks 
srl

BryceJ wrote:

Ran into an issue with IRQ balance after updating to trixie
 

Late to the party but currently best practice is not to use irqbalance. the balancing is now managed internally in the kernel so adding a second balancer can confuse things.... but it worked here....

i have upgraded to trixie 6.12.113 kernel and jitter now is much worse..
Any updated instructions for improving jitter?
I think every kernel update make things worse.

I have found that 7.x kernels are better than 6.s kernels jitter wise.

How to install 7.x to debian 13 from backports repo ?

Unfortunately,  I don't know about Debian, I use Mint
and build kernels from source.

PCW wrote:

Unfortunately,  I don't know about Debian, I use Mint
and build kernels from source.

The procedure is the same for both.