How do I hack CPU usage precision (from getrusage) by building armv7 image

[Crossplatform Vlad]

Build logs:

https://paste.armbian.com/ivisomijaq

Please take a look on CPU usage precision on the new kernel for x64 and arm64. It is 1 milliseconds.

But kernel 6.13 on armv7 still provides 4 milliseconds of precision.

 

How do I tune CPU usage precision on image build?

 

[Crossplatform Vlad]

Seems Ubuntu from Canonical uses a customized kernel where CPU usage precision is improved. 
But Debian and Armbian use previous CPU usage precision settings?

 

[laibsch]

Can you point to a patch or some code from Ubuntu how they achieve that?

[Igor]

On 5/25/2025 at 1:16 AM, Crossplatform Vlad said:

But Debian and Armbian use previous CPU usage precision settings?

Did you compare those things on the same hardware (architecture)?

[hexdump]

maybe its related to the kernel config parameter CONFIG_HZ_xyz - depending on which of them is set for the different kernels?

[Crossplatform Vlad]

On 5/26/2025 at 10:26 PM, Igor said:

Did you compare those things on the same hardware (architecture)?

Good question. I did compared predefined default kernels from canonical (6.8) and debian (6.10) on the same hardware twice:
1) qemu arm64 emulator on x86_64 host with the same parameters. lines 6 and 7 on the screenshot
2) qemu x86_64 emulator on x86_64 host with the same parameters. lines 4 and 5  on the screenshot

 

[Crossplatform Vlad]

On 5/26/2025 at 10:38 PM, hexdump said:

maybe its related to the kernel config parameter CONFIG_HZ_xyz - depending on which of them is set for the different kernels?

Seems it is an option for timer resolution in the measurement of duration. But the topic is about CPU usage precision returned by getrusage function. it is hardcoded interval of time when kernel updates CPU usage of particular thread/process. 

Edited Friday at 11:47 PM by Crossplatform Vlad