Script for automated testing of computer latency.

Maybe it's useful to report my latency experience. I am sharing only some overall figures, not the entire logs, sorry for that ...

Hardware: Asus M2NPV-VM, Athlon X2 BE-2350, Matrox Millenium II, 4GB memory, SSD.
Software: Debian 10 with RTAI.

The latency (base thread ) observed by running the script for more than 12 hours was around 14us.

But the more interesting result was obtained by running over ssh and disabling graphical interface completely on the linuxcnc machine ( runlevel 3) : under 7us for 12 hours test. This time I have only used latency-histogram and I have generated relevant load for a "runlevel 3" situation.

Disabling graphical interface is definitely not a new thing.  I wonder if anybody is  successfully using such setup for real action.

msurmei wrote:

Maybe it's useful to report my latency experience. I am sharing only some overall figures, not the entire logs, sorry for that ...

Hardware: Asus M2NPV-VM, Athlon X2 BE-2350, Matrox Millenium II, 4GB memory, SSD.
Software: Debian 10 with RTAI.

The latency (base thread ) observed by running the script for more than 12 hours was around 14us.

But the more interesting result was obtained by running over ssh and disabling graphical interface completely on the linuxcnc machine ( runlevel 3) : under 7us for 12 hours test. This time I have only used latency-histogram and I have generated relevant load for a "runlevel 3" situation.

Disabling graphical interface is definitely not a new thing.  I wonder if anybody is  successfully using such setup for real action.
 

This is interesting! Could you kindly post the .hw file? That's a summary of hardware, actual test run log (I am particularly inerested in the no of glxgears and firefox) and just the latest two lines of the very latency data. How did you manage to see what's going on? remote X-server?
Greez
chris

seuchato wrote:

rdeeming wrote:

This change to the latency-histogram line 696 messes up the current script feeding off "secs" for reporting to the hw log file.

Actually the part for latency in the xyz.hw file is just the last two lines of the xyz.lat file that contain latency data. (like that, servo and base thread are covered). I'll have to check if at all and if, how that change affects logging of latency data. Hope I can do it soon.

If this turns out to be a problem, can I suggest grepping for both "secs" and e.g. "us sdev" to keep the script compatible with both old and new versions of latency-histogram? "secs" should get lines from the old version, and "us sdev" should get lines from the current version. 'grep -e' should do the trick?

a solution is a solution. As an engineer, I could not oppose to a solution :-]

Glad to hear it!

Grin