7i96 issue read/write

One quick question. IIRC I configured the original NIC via GUI. Is it necessary for me to set IRQ coalesce to zero. If so, I think I need to set: sudo ethtool -C eth0 rx-usecs 0 as per tommylights instructions. Is that right?

Does this config persist when setting are made graphically or is it best to resort to editing the etc/network/interfaces file. Tommy does warn not to do both so just want to check. Also I am not clear if this just relates to onboard intel ethernet chips or if its necessary here with a standalone pci card. I can run some pings tomorrow to see how it is. Initial tests are positive.

sudo ethtool -C eth0 rx-usecs 0

is temporary (until a reboot)
The interfaces file setting runs that command at boot time

Okay. Thanks, for that. I thought that was the case. Ill remove the gui config and shift config to the etc file then since I assume the is no graphical way to configure these interrupt settings.

Some more testing today. Unfortunately this has not resolved the issue. Given that this is happening on two different machines with independent 7i96 cards I can at this point only assume it is the system itself or power related. Each cnc machine has an overspecified 5v psu (meanwell) solely for powering the 7i96. Voltages seem very stable and current is no where near what is required by the board. I have to at this point assume it is related to the comp as I'm running an early release of 2.9. As next step ill try updating this to a fresh install when my new ssd arrives.

Ping times are long even with irq settings enabled, circa 400ms in cases. I think I would benefit from a different kernel regardless. Some work to do...

I have changed a bunch of things over the past couple of days, moved up to Debian 11 building from Git, but found that it produced much higher servo jitter on my machine. On Stretch, Buster my machine is solidly under 5000 for servo, on 11 it jumped up to between 25,000 and 30,000.

I ended up wiping the drive and moving to Buster, my 2.9 configs would not play nicely with 2.8.4 so I upgraded to 2.9 and am able to start both machines now. I am running dry run tests on the router as we speak, with lower than normal velocities. I will bump this velocity up in stages over the next few days to see if I encounter this same error.. For reference the current Buster install is on 4.19.0-23-rt-amd, I am testing the Intel NIC with recommended IRQ coalesce settings applied.

Note that normal latency test measurements are not really significant
With Mesa Ethernet hardware (unless they greater than 100 or so usec)

Much more important is network latency, checkable with:

sudo chrt 99 ping -i .001 -q 10.10.10.10

(let run for a minutes or even hours) and stop with control C