Installing LinuxCNC On HP T610 Thin Client

tommylight wrote:

rodw wrote:

tommylight wrote:

Then since you use QtPyVCP a lot, see here:
www.qtpyvcp.com/install/bullseye.html
Don't mind me laughing Tommy. Joco wrote those instructions! 

Yeh, I worked out Debian 11 about a week before the official release (when it was testing) and reviewed Joco's instructions and he updated them to make it more complete. Those instructions should include all the dependencies for qtvcp as well as qtpyvcp. I am not loving debian but at least it works! 

Joco wrote:

In the case of this machine i suspect your issue is kernel/distro related.  When i did my initial work i tried debian 10 and ended up on mint 19.3 for a very good reason.

i have not done any tests on Debian 11(bullseye).  I am not surprised it performs poorly. If you are serious about needing bullseye then you might need to look at custom kernel compiles.  

Ok dude, what magic are you using ?

Installed Mint 19.3, have a 4.9-13-rt kernel from debian and all my latency is over 300us using your script, around 280us using the original script......Yep same kernel cmdline options.No sound or pc_speaker modules, tried all my ususal tricks

Even my 4.19-106-rt kernel gives crap results

The only diffs are I have the later BIOS on my T610 and memory is a single 4GB module. The single memory module is the only thing I can think off that might be an issue.

No better when forcing the GPU memory from 128 to 64.

When you run the attached script when trying to do a latency test what cpu frequency vales are you seeing?

 

Installed from the Buster ISO and max latency dropped so much so that’s it’s a tad off being also passing with a 1ms servo thread and a 25us servo thread.
Running no base thread and latency tests pass with a lot of room to spare.
I have another 4GB stick of ram coming in and I’ll see if that makes any difference.

what kernel are you running now?  I know from my experiments that there is an older RT kernel that drives the cpu's at max clock speed without having to do anything special.  That was why I was asking about the results from that script.  It monitors the CPU MHz.  As under the kernel version you were using on Mint you do need to be running the CPU hog script on the second CPU core. Not doing this will result in the crap latency you were seeing.

J.

 

Std one that's on the Linuxcnc ISO. Tho I added Caja for my file manager, Mate Terminal and the gvfs bits that are missing. Caja & Mate terminal pull in Mate libs. I just prefer those 2 over the xfce4 stuff. Oh yeah I also pulled in the xfce4-power-manager package.

Didn't make a difference whether I ran the CPU hog script or not. I just don't get it. Two machines that are the same (apart from BIOS) and getting different results.

Tho I noticed in your testing you have no parallel port & my machine reports a parallel port. All parallel port modules load.

I worked on it for while. Used my Mint respin, then used a standard Mint ISO and followed tommy's instructions, including the kernel mentioned in that thread. Snd modules were blacklisted in all tests, different combos of kernel command lines.

Joco wrote:

When you run the attached script when trying to do a latency test what cpu frequency vales are you seeing?


 

Joco,
I hope you will not mind me, adding this to my scripts for latency testing?
Have to figure out on where and how it is usefull and what to protocol.
Greez
Chris

ALittleOffTheRails wrote:

Std one that's on the Linuxcnc ISO. Tho I added Caja for my file manager, Mate Terminal and the gvfs bits that are missing. Caja & Mate terminal pull in Mate libs. I just prefer those 2 over the xfce4 stuff. Oh yeah I also pulled in the xfce4-power-manager package.

Didn't make a difference whether I ran the CPU hog script or not. I just don't get it. Two machines that are the same (apart from BIOS) and getting different results.

Tho I noticed in your testing you have no parallel port & my machine reports a parallel port. All parallel port modules load.

I worked on it for while. Used my Mint respin, then used a standard Mint ISO and followed tommy's instructions, including the kernel mentioned in that thread. Snd modules were blacklisted in all tests, different combos of kernel command lines.
 

ALittleOffTheRails

• Kernels do have a major impact. It is sometimes a lot of trial and error to find the best one.
• "Two machines that are the same (apart from BIOS) and getting different results." I still have a dell T3500 that I can't get decent latency values with, while someone else got very good results. I did almost everything on this robust piece of hardware: Ram, CPU, Disc, Bios, Distro, Kernels, window manager, desktop, boot cmd line, you name it: no joy, not at all. A friend of mine has another T3500. It is still serving his home. Once decomissioned, I was promised to get it for some testing.
• RAM: next weekend I might get time to do some (limited) comparison of a known good (imho) machine (Dell 980) that I can run with 1, 2 or 4 x 4GB Ram. I will edit this post to add the link with the results. Please add your findings with 1x and 2x 4GB Ram. A remark: I remember having mixed results with mixed Rams.

Greez
chris
 

Yep I know about kernels. I spent a fair bit of time testing before I released my Mint respin. Which seemed to work well, even on the Odroid H2 plus once I added the Nic module.
For the T610 it would seem that the kernel on the official Linuxcnc 2.8.2 iso gives best results, from my experience.
Anyways if more academic than anything else at the moment, as I still have some work left to do to get the mill back to work before I start on the lathe, tho I’m kind of in 2 minds about that.
So no more collecting computer hardware for the time being as I need some spare cash to get some fibre glass supplies.