Advanced Search

Again, thank you, but again, i never ever used isolcpus in my life, yet i have over 30 machines running LinuxCNC and Mesa hardware happily, every day, on mostly industrial machines, for over 10 years.
Still, i understand that sometimes isolcpus and some other tuning can help.
P.S. that 30 above does not include over 15 PC's and laptops i own, all with LinuxCNC and used on machines with Mesa boards, and therein lies my dilemma, i have plenty of laptops that run LinuxCNC out of the box, but i can not advise on getting the same laptops as i had two of the same make and model where one was OK and the other was utterly useless. And i use USB-to-LAN adapters and switches, but i never ever advise others to do the same thing, as i am fully aware of the consequences, if something fails.

I did not say to switch slots, although that might help on old PC's or if the memory pins were touched during installation, i said to test each module, one at a time, and that is still valid advice, had new 16GB module fail inside a week lately, out of 6 modules one failed, and the PC will boot with it and work till starting new apps does not work anymore and the PC reboots on it's own.
Sent it back for replacement, got it back with the note "it works fine", now i have to call tomorrow (well today as it is 5:15AM here) and explain to them that booting windows does not work as a memory test.
I did test it, on 4 motherboards, on 3 it works and memory test fails miserably, on one board it will not boot at all, even after BIOS reset and XMP/EXPO disabled.
The other things that might cause such issues, power supply (hard to check without replacing, needs a scope), and in your case might also be the motherboard as A520 are the low tier for AMD, so signal integrity for the memory controller might be compromised, although AMD is pretty damn good at fixing errors.
But first, disable XMP/EXPO in BIOS and test.

  The original computer I was trying to use was built off an ASUS A520m. I did do a RAM check. They both reported working fine. I have 2 8gb sticks, running together for a total of 16gb, details shown in the screenshot. Why would switching RAM slots matter? 

Ooops, I saw the pings but missed the disconnect window. So to my mind, the only things that could cause this is faulty hardware, crappy network cable or a flakey power supply on the Mesa card. The Mesa power can be eliminated at this point. Time to follow PCW's advice to isolate that piece of hardware is causing it.

But the ping times are also too slow.

None of my suggestions are out of scope  of a  mesa card and earlier screen dumps showed no isolcpus was set so a lot of work to do there 

I am glad as i ran out of options a long time ago as most of it did not make sense!
What Asus?
If it has two or more sticks of RAM, yank one out and test, then yank the other and insert the first one, test again.

Rod, this would have been much more useful in the "driver boards" or EtherCAT or both sections, not here as OP has actual networking disconnecting, same as physically removing the cable.
I really appreciate you trying to help, but i really do not like when you point members of this forum and whoever reads this in the wrong direction.
This is the second time i am mentioning on this topic that none of that tuning will help with a disconnected link.

Forgot to mention it installs the realtek dkms drivers so your NIC's should be fine as Linux will install the correct drivers but it may select the first NIC for Ethercat. The automated tuning on my tests is about 50 % better than any of my  manual tuning attempts.
typing sudo verify will generate a PDF report card.

Typing sudo MrRodW-tuner --help will list some available options including selecting a different NIC but a bit untested right now. Sorry, it does not know German but I did try to treach it Italian at one stage but I could not test it here in Australia.

My solution via an ISO here
mrrodw.au/en-usd/products/v298deb
This gets the new packages from Linuxcnc-ethercat for both the Hal driver and the Ethercat master.
I also have a ISO for 2.10 which is essentially the same but picks up the new scurve (jerk limited) features etc.
Once the ISO is installed, a tuner program runs and selects the best NIC for Ethercat and completely installs Ethercat and reboots to lock it in.
I'm currently modifying the tuner so you can run it outside of the ISO process and select a specific NIC but you should not need to do this for a new build.

nice work guys!

If its working, I think you should add a PR of the driver, not the proof of concept hall, inin anf\d GUI extensions

Here is a checklist of the key optimizations required for a `PREEMPT_RT` real-time kernel on Debian, specifically tailored for network performance on interface `eno1`.

### 1. BIOS/UEFI & Hardware Level

* Disable **Intel SpeedStep / AMD Cool'n'Quiet**
* Disable **Global C-States** (C1E, C3, C6, etc.)
* Disable **Intel Turbo Boost / AMD Core Performance Boost**
* Disable **Hyper-Threading (SMT)** *(highly recommended for strict determinism)*
* Disable **System Management Interrupts (SMIs)** / Advanced Power Management

### 2. Boot Command Line (`grub-customizer`)
lastcore = the last CPU you have. say its 3 on a 4 core CPU but yours may be different
* Add `isolcpus=3` to isolate specific CPU cores for RT tasks
* Add `nohz_full=3` to enable full dynticks on isolated cores
* Add `rcu_nocbs=3` to move RCU callbacks off isolated cores
* Add `processor.max_cstate=0` and `intel_idle.max_cstate=0`
* Add `pcie_aspm=off` (disables PCIe Active State Power Management)
* Add `skew_tick=1`

### 3. Network Interface (`eno1`) Optimizations

* Disable **Adaptive Interrupt Moderation** (`ethtool -C eno1 adaptive-rx off adaptive-tx off`)
* Set static interrupt coalescing to 0 (`ethtool -C eno1 rx-usecs 0 tx-usecs 0`)
* Disable offloads causing latency spikes:
* Generic Receive Offload (`ethtool -K eno1 gro off`)
* Large Receive Offload (`ethtool -K eno1 lro off`)
* TCP Segmentation Offload (`ethtool -K eno1 tso off`)
* Bind `eno1` queues to the single dedicated, isolated CPU core

### 4. Operating System & IRQ Tuning

* Increase the real-time priority of the `eno1` threaded IRQ handlers (using `chrt`)
* Set the `eno1` IRQ affinity mask to use isolated core
* Set the system power management policy to performance (`cpupower frequency-set -g performance`)
* Disable the `irqbalance` daemon (`systemctl disable irqbalance`)

 

Well guys, I switched to a humble dell optiplex 5050 and the everything works perfectly. 
nuclear option. It still bugs me I can’t figure out what’s wrong with the ASUS. But I’ll take the ‘win’ for now…

Do you have anther PC or another Mesa card to see which device is dropping the connection?

Guys, literally as I was typing that this only happens with MPG stimulus, the computer crashed as I was launching LinuxCNC in the background! I completely unplugged all encoders and launched LlnuxCNC again, and it crashed again. 
This is no longer an MPG issue! Its a random computer issue I cant figure out. 
The crappy part is every component on the computer is NEW, and tests fine. There has got to be some BIOS setting I am missing. 

At least I can stop chasing my tail on the MPG. Before anyone asks, the crash logs are blank, no reported errors.  

As a point of interest, there is still something screwy going on with the computer being able to find the 6i25...

Hi Mueke,

Der ethercat hal Driver wurde komplett überarbeitet.
Beachte diesen threat:
www.forum.linuxcnc.org/ethercat/58103-et...m-jitter-fix?start=0

Das GitHub repo ist hier:
github.com/linuxcnc-ethercat/linuxcnc-ethercat

Wenn du Hilfe benötigst, sag Bescheid