Mesa 7i96S - Error finishing read iter ...

Hello all,

I am using a Mesa 7i96S coupled with a mini PC for LinuxCNC but I am facing what seem to be latency problem.

Sometime, I get a error at startup complaining about a abnormal task delay of over 1000000.

Sometime, it start and work normally but after a few minutes the machine stop with a error like "Error finishing read iter : xxxx".

I really don't know what to look at ?

Here are data that I think will be required to help me :

Have you disabled all power saving options in the BIOS?
(C states > C1, turbo modes etc)

Can you use a later kernel? (6.3 or later are usually better)

If nothing else works, you can try increasing the servo thread period to 2 ms
(2000000 ns)

PCW wrote:

Have you disabled all power saving options in the BIOS?
(C states > C1, turbo modes etc)

Can you use a later kernel? (6.3 or later are usually better)

If nothing else works, you can try increasing the servo thread period to 2 ms
(2000000 ns)

In BIOS I have exactly disabled :
- VT
- Turbo Mode
- SpeedStep
- C-STATE

I will try to use a later kernel so.

You think my issue is the max Jitter ?

Are my hm2.* values good ?

Interpreting the max jitter values requires the CPU clock speed
(as those value are in CPU clocks)

Assuming a ~2 or ~3 GHz CPU those numbers look OK

(but I suspect they were taken before you got an error...)

PCW wrote:

Interpreting the max jitter values requires the CPU clock speed
(as those value are in CPU clocks)

Assuming a ~2 or ~3 GHz CPU those numbers look OK

(but I suspect they were taken before you got an error...)

It's an Intel N100 based fanless PC.

Honestly I don't remember if it's before or after the crash.. I will check again.

While I have an expert at my disposal Can you confirm that I do understand the basic of LinuxCNC :

- If my axis has a scale of 200, it mean my stepper need 200 pulses per mm of travel.
- If my step lenght and step space are 5000 ns, it mean 5uS so it can theorically pulse 100.000 times per second.
- If I need my axis to move at max 40 mm/s, it make 200 * 40, so I only need 8000 pulses per second ?

Lastly, what is the servo thread period ? As it seem way slower than the pulse rate requirement. I don't understand it.

Thank you.

A scale of 200 means 200 pulses (or encoder counts) per machine unit

A step length of 5000 ns and step space of 5000 ns would limit the peak step
rate to 100K steps per second

Yes, 40 mm/sec with 200 steps/mm is 8000 steps/second

The servo period determines how  often LinuxCNC updates positions
and in the case of most controlled hardware, how often the
velocity is updated.

In the specific case of Mesa hardware stepgens, the stepgen step rate
(which can be as high as 10 MHz) is adjusted ever servo thread.
 

A couple other things that help network latency are

1. If you have an Intel Ethernet chip, turn off IRQ coalescing
(man hm2_eth)

2. Pin the Ethernet IRQ to the last CPU ( setting the Ethernet  IRQs smp_affinity to 8 does this on a 4 core CPU)

 

I switched to kernel 6.5 but don't really have improvement.

I checked the NIC in that mini computer, it's realtek chip.

What is strange is that error happen out of nowhere while idle. As you can see on this screenshot, I have turned on the machine at 18h20 and have homed the axis. Then I have leave the machine alone while eating and when I come back I see that error has happened at 18h49 while doing nothing.

 

When you get the error, do you have a red LED illuminated on the 7I96S?

Did you try a 2 ms servo thread?

PCW wrote:

When you get the error, do you have a red LED illuminated on the 7I96S?

Did you try a 2 ms servo thread?

I am not sure but it's possible that there is a red led, I have to check again. Does it mean a specific problem ?

Not yet with 2ms thread, is there any negative effect ? I guest if my machine is quite slow it should not induce tracking error ?