Mesa 7i92 Read Errors

Hi all,

I just changed the computer on my machine to an Asus PN50 from a Intel NUC 10 due to graphics issues.
While the graphics issues are gone, and the overall latencies look phenomenal, I am now getting the "Error Finishing Read", usually after less than 1 minute of running.
When it first ran, I checked the read and write latencies, which came out at around 10000 (without errors).

I already tried changing the Ethernet cable (currently a shielded CAT6), and the power supply should also not be any issue.

I have an Mesa 7i92 attached to a 7i77, 7i85s, and 7i73.
Also, I am running a custom lubuntu install with linux kernel version 5.18 with the RT patch applied and configured.

Thank you for reading.
 

Hi tommylight,
I just ran two ping test to the 7i92, and they are well within the timeout.


Standard
Flood Ping ()

Hi timmylight,
Thank you for the suggestions, we already ran the latency histogram with 10 glxgears, for a longer period than the screenshot above showed.
We will run it again tomorrow because we decided to rebuild the kernel to install the latest RealTek Ethernet driver.

The power supply is a din rail mounted 5V 3A power supply from MeanWell powering only the MESA cards and some optocouplers.
As for interference, it is extremely unlikely as we are using a shielded CAT6 cable that is run parallel to only 24V lines.

 

Hi tommylight,
The issue happens with and without the drives are energized.
Furthermore, I already disconnected both cards from the 7i92 and the issue persisted.

And while you are correct in the DYN-4 being connected to an analog signal from the 7i77 with velocity mode since I have linear scales on 2 axes, with the 3rd axes being controlled via Step/Direction from the 7i85, the drives are not enabled due to the mesa card losing communication well before I enable them.

1. What is your servo thread period?
2. Have you checked the tmax values when running LinuxcNC?

halcmd show param *.tmax

To interpret these values, the CPU frequency is also needed.
(at the tmax values are in CPU clocks)

I should mention that you can eliminate power supply issues
(and throw suspicion on timing issues) if  the /INIT light is
illuminated on the 7I92 after you get a read error.

 

Hi PCW,
Thank you for the points to look at.

I just tested the read and write tmaxes and they jump up to 1ms for some cycles, with my servo thread cycle time being 1ms as well.
Also the INIT light stays on after the error ocurrs, confirming that it is a timing issue on the ethernet interface.
I’m currently in the process of chasing down why it it jumps up sometimes.