0

I'm doing a study to compare the linux default kernel with the kernel patched with preempt_rt. I used a lot of tools (stress-ng, T-Sui, CPU burn, cycletest) to check the effectiveness of the patch. But I can see that the default kernel does already provide a method to change the tasks priority to a "real-time-task". Therefore, all of the stress-tests show a similar result: my application is always performed as real-time task without any strong delays or something different surprising...

I know some differences that preempt_rt causes (like interrupts moved to thread context, less non-preemtive code in kernel) but how do I test them?? I've seen that the normal stress-tools show almost no impact on my real-time-application... Is there any tool to execute a kind of "high-non-preemtive-kernel-code" or that triggers a huge amount of interrupts?

MaMa
  • 31
  • 4
  • You can always see if [scheduler-tools](https://github.com/scheduler-tools) fits your purpose, specially [rt-app](https://github.com/scheduler-tools/rt-app). An option to test your application is to generate artificial interrupts with higher priority than your application and observe the impact in terms of schedulability and jitter. – CRM Apr 05 '20 at 10:20
  • @CRM How do I generate artificial interrupts? – MaMa Apr 05 '20 at 10:56
  • In [this](https://stackoverflow.com/questions/27865075/how-would-one-go-about-generating-artificial-interrupts-in-the-linux-kernel) post you may find different alternatives to do that. – CRM Apr 05 '20 at 13:02
  • Have you managed to succeed with this? – CRM Apr 06 '20 at 07:06

0 Answers0