When searching for term 'rt_preempt' there are past questions on the topic of reducing jitter on the Codesys Linux runtime, running on a Debian-based OS. I found these suggestions:
Use of the RT_PREEMPT kernel patch
Ensure stable CPU clock
Use of CyclicTest tool for diagnostics
To further improve performance (reduce jitter), are there additional suggestions from Codesys or the others in the community? For example:
Debugging kernel latency
Building a smaller/"lighter" kernel
CPU isolation/pinning
others?
It would be ideal if Codesys could compile a whitepaper with their most common practices and suggestions to help Linux runtime customers get the best performance possible. But for now, perhaps we can work together in the forum to share/learn.
👍
1
Last edit: adverb 2020-07-10
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi all. cpupower utility was helpful in setting governor to 'performance' mode in Debian, which appears to help stabilize/lower cyclictest numbers, even while under stress-ng loads. Any tips for debugging outliers? I mean to say, debugging a missed latency target, by tracing kernel to find the cause?
👍
1
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hello Codesys Community,
When searching for term 'rt_preempt' there are past questions on the topic of reducing jitter on the Codesys Linux runtime, running on a Debian-based OS. I found these suggestions:
To further improve performance (reduce jitter), are there additional suggestions from Codesys or the others in the community? For example:
It would be ideal if Codesys could compile a whitepaper with their most common practices and suggestions to help Linux runtime customers get the best performance possible. But for now, perhaps we can work together in the forum to share/learn.
Last edit: adverb 2020-07-10
There is no release planned, yet, but it is on our list!
more posts ...
Hi all. cpupower utility was helpful in setting governor to 'performance' mode in Debian, which appears to help stabilize/lower cyclictest numbers, even while under stress-ng loads. Any tips for debugging outliers? I mean to say, debugging a missed latency target, by tracing kernel to find the cause?