Arch Linux

I've copied the settings from Clear Linux. They probably have good reason to choose these settings for a general purpose kernel.

Clear Linux isn't general purpose kernel. It's designated for VMs and cloud. Moreover those settings are meaningless if you want to achieve Clear Linux performance as it needs their whole toolchain and rebuilding userspace with it.

I just checked this and only Fedora has 1000Hz while Ubuntu/Debian/OpenSuse defaults to 250Hz. Who has good reason and who hasn't? I linked discussion where people report more heating and power usage under 1000Hz which I can confirm myself. Searching web brings more of the same.

I don't like that things are copy-pasted based on pure belief of "good reason". We need testing, benchmarks and discussion instead. I remember enabling intel-iommu thing which I could instantly report that it won't work.

As I wrote we have linux-zen package which anyone concerned about those settings can use.

> Clear Linux isn't general purpose kernel. It's designated for VMs and cloud.

It's also for IoT devices (small Atoms and such), with the kernel I looked at being "optimized for bare metal usage" which makes it general-purpose enough.

> Moreover those settings are meaningless if you want to achieve Clear Linux performance as it needs their whole toolchain and rebuilding userspace with it.

This is blatant bullshit. The changes to the kernel do not require changes to userspace. Kernel changes won't get us on par with Clear Linux, but it can help.

> I linked discussion where people report more heating and power usage under 1000Hz which I can confirm myself.

If you had actually read what you linked to you would have noticed the bug is about lower power use when using 1000 Hz.

> I don't like that things are copy-pasted based on pure belief of "good reason".

Our previous settings didn't have any better reasoning than "it wasn't noticeably bad", either. However, I believe that Intel has professionals working on Clear Linux that do know what they're doing.

> We need testing, benchmarks and discussion instead.

We do. Actual benchmarks would do a much better job at convincing me.

> As I wrote we have linux-zen package which anyone concerned about those settings can use.

linux-zen is irrelevant, as its config will remain close to the normal linux package.

> This is blatant bullshit. The changes to the kernel do not require changes to userspace. Kernel changes won't get us on par with Clear Linux, but it can help.

Here's benchmark between generic Ubuntu kernel (250Hz) and lowlatency one (1000Hz):
https://www.phoronix.com/scan.php?page=article&item=linux414-lowlatency-liquorix

"The generic Linux 4.14.13 had 15 wins (35%), the Linux 4.13 generic kernel 10 wins (23%), the Liquorix kernel with 9 wins, and then the low-latency kernel with just 8 wins."

Is that what you call "blatant bullshit"? Obviously Clear Linux will win above benchmarks easily but this https://gcc.gnu.org/wiki/AutoFDO is their secret sauce not CONFIG_HZ=1000 or CONFIG_NO_HZ_FULL=y.

> If you had actually read what you linked to you would have noticed the bug is about lower power use when using 1000 Hz.

I wrote "see comments":

"300 Kernel tick rate used by Arch linux as a culprit was a red herring. The root cause issues were also present in the 250 Hz and 1000 Hz kernels" - which invalidate your answer.

"I did some tests with changing CONFIG_HZ to 250hz like Ubuntu and openSUSE Tumbleweed and I get better/lower power use than with 1000hz and noticeably lower than Arch 300 Hz." - which confirms my point.

> linux-zen is irrelevant, as its config will remain close to the normal linux package.

It's relevant because if you want to do some experiments you can do it there.

If your answers for my arguments are "beliefs" and "bullshit" then I don't care anymore about convincing you and I'm done. Feel free to close this.