updates 2.9 Bookworm

Serg452 wrote:

Key at buildbot2.highlab.com/debian/dists/bookworm/
Current: 2.9.0-88-g6ded497531

Lcvette wrote:

I wish the bookworm latency issues would get dialed in and sorted.
 
I've managed to dial my latency down to +/-0.04ms with ethtool. Contrary to the infamous hardware-irq-coalesce-rx−usecs 0 documented (this 0 didn't worked for me, made things worst in fact), some tuning/testing is required. 

Using ethtool --coalesce
$ sudo apt install ethtool
$ sudo ethtool -c eth0
( take notes of your rx-usecs and tx-usecs defaults. Mine were 100 and 1000 respectively)
( start Linuxcnc and have it connected to the card)
( Time to dial/tune)

$ sudo ethtool -C eth0 rx-usecs 50
$ sudo ethtool -C eth0 tx-usecs 500
$ ping ip.ip.ip.ip
....
$ sudo ethtool -C eth0 rx-usecs 25
$ sudo ethtool -C eth0 tx-usecs 250
$ ping ip.ip.ip.ip
....
( Repeat tuning until you get satisfactory latency and no lcnc task/axis pos errors, etc. )
( Once everyone's happy, save those in your interface conf. )

$ sudo nano /etc/network/interfaces

***********************************************
      # The loopback network interface
      auto lo
      iface lo inet loopback

      auto eth0
      iface eth0 inet static
      (... ... static address stuff)
      hardware-irq-coalesce-rx−usecs 25
      hardware-irq-coalesce-tx−usecs 250
***********************************************

$ sudo systemctl restart networking

 

thanks,  I will give that a try and report back!

get an error trying this when i enter the tx line and don't get much info from sudo ethtool c enp0s25

thoughts?

 

You cannot set TX usecs on some hardware

If you watch hm2_[card].0.read.time as you adjust the rx-usecs value
You will find the time time gets longer for any rx-usecs value other than 0.

This is not surprising since the rx-usecs value is how long the host should
hang in the Ethernet interrupt service routine before exiting. Waiting make sense
if you are expecting another packet but this is not the case here.