joint 3 following error

Hello. I have a problem. Sometimes the error “joint 3 follow error” pops up. The problem is not stable. The problem appears to occur when the job is restarted. I tried to track the error values ​​in the Halskop and see that the error values ​​are a maximum of 0.07 millimeters. I spent several hours testing the machine and the error did not appear. The day before I managed to see an error in the terminal, and I will show it in the photo. I use stepper motors, so I
have a simple step control system with no feedback.I don't know how to make the error appear again. I did not change the configuration files. I have no reason to change the values ​​since I can't catch the error and understand the reason. I suspect that the reason may not be in the settings, but in linuxcnc or in the operating system, or in the network card drivers...
I would be grateful for any advice.
And I apologize for my English. I'll add more information.
On the day I saw the error, I could press the power button on the machine and repeat the task. And when I repeated the task twice, the error appeared consistently. I saw the error 10 times that day.
The next day I restarted the computer, did not change anything, and now the error does not occur. But I'm worried that the error will appear again. what can you recommend to fix the error? Which tool should I use?

try to set ferror 50 and min_ferror 5 ... than see if work .... maybe simply ferror too low for your latency .... and maybe ferror of 50 was only few 0,00x mm ....

Can you post your hal and ini files?

Those are relatively huge errors so not likely latency related
(at 50 mm/sec a full servo period (1 ms)  late would only
result in a  _apparent_ following error of .050 mm

Good afternoon. 
I thank you that my problem is interesting to someone. 
I don't have the configuration files at the moment and will of course attach them later. 
For now I will show the result of replacing the network adapter. At the beginning of the topic I used Realtek. 
Using the materials of the article 
forum.linuxcnc.org/9-installing-linuxcnc...dware-eg-realtek-nic 
I installed the patched kernel. I also tried the r8168-dkms driver, but it doesn't give any tangible results. I thought the culprit was the network adapter, so I tried Intel. But intel turned out to be worse in terms of ping time. Also, the exchange time is very strange, and looks like a see-saw.  And this saw is affecting the servothread time . On risinuk the machine is stopped and at rest. 
I will also attach a histogram of my tests. The histogram does not look bad. 
But! We are looking at the histogram for a machine that does not move and in this case there is no network communication and there are no algorithms to control the machine. 
I've also attached a picture with a realtek network. And the halskope realtek looks very good until the machine starts moving. While the machine is moving, the servothread timing graph has very large spikes, very high needles. 

 

I found the configuration files. They are not up to date, but I think not much has changed since then.
But later I will post the actual version too.

Let me add. 
I believe that the servothread time should be stable. The time should not depend on whether the machine is in motion or not. But in my case servothread time starts to have spikes exactly during motion. 

Can you print the result of

halcmd show param *.tmax

after running LinuxCNC for a while?

And also what is your CPU frequency?

The sawtooth read time is bizarre, I have never seen anything like it...

Also on Intel MACs you must disable irq coalescing to get consistent ping times

My CPU: Intel Core i3 3240, 3.4 GHz, 2 cores
My mother board: gab75m-d2v
RAM: 2x2Gb
SSD

How to disable irq coalescing?

Once I get to the workshop, I'll do the tests you recommended.

Perhaps you have more thoughts on how to set up the experiment.
Oscilloscope

I'll try to be back in a few hours once I get to the garage

You disable irq coalescing with ethtool
These is an example in the hm2_eth man page

To disable coalescing temporarily for a test (it will revert to enabled on reboot):

sudo ethtool -C [ethernet_device_name] rx-usecs 0

Hey, everybody. Thanks for the reply!Last night, together with my comrades, the problem was solved in this way. 
sudo ethtool -C [ethernet_device_name] rx-usecs 0
This was suggested by a member of this forum - PCW. And also in "telegramm" messenger Maxim also suggested to do it this way. 
In addition, before 
sudo ethtool -C [ethernet_device_name] rx-usecs 0
Maxim suggested installing
sudo apt-get install ntp
But the ntp installation conflicted with systemd-timesyncd and I uninstalled systemd-timesyncd. 
After that, sudo apt-get install ntp succeeded. 
This may have had some effect, I'm not sure, but the saw on the graph continued to remain. 
The ntp systemd-timesyncd was written about here. 
forum.linuxcnc.org/38-general-linuxcnc-q...or-problems?start=20
After applying
sudo ethtool -C [ethernet_device_name] rx-usecs 0
the problem was solved. And I am attaching the picture. I managed to capture on oscilloscope the moment of applying the command in terminal. There is no doubt, the effect is definitely there!I have no problems at the moment.

I am grateful for the help!I feel obliged to leave this message here. 
By the way. intel ping has gotten better.
I have made the decision to use intel. I will be testing the machine.
I hope everything will be fine.