Fedora Gets An Unofficial Kernel Based On Clear Linux

[Candy]

For some reason this reminds me of http://funroll-loops.teurasporsaat.org/

Don't reply if you don't understand the issue.

[wizard69]

Honestly no thanks at this point, my Ryzen Mobile system just became stable! By the way a very big thanks to everyone focused on resolving the bugs in the kernel, libinput or where ever they are, to get Ryzen based laptops to wake up correctly. In this case i can say that a Ryzen based HP ENVY and Fedora are ready for prime time.

By the way this doesn't mean that Fedora or any other distro should ignore performance!!! Rather id rather see it delivered in increments that stress reliability. For the most part im happy with performance today on this generation of hardware. Better never hurts but not at the expense of proper function.

[Venemo]

Kudos to Intel for making Clear Linux.

However, thumbs down to them for not upsteaming their patches. Instead of making a "clear linux kernel" for Fedora the work should have went into upstreaming those patches and then everyone would benefit.

That being said the Clear Linux performance story is not just about the kernel. They basically added various optimizations everywhere and I would imagine most of their performance is not related to just the kernel.

[onicsis]

Phoronix: Fedora Gets An Unofficial Kernel Based On Clear Linux

While the kernel configuration is just one part of Intel's Clear Linux optimizations for their performance-oriented distribution, a Fedora user has taken the liberty of spinning a Fedora kernel build based upon Clear Linux's kernel configuration...

http://www.phoronix.com/scan.php?pag...l-Clear-Kernel

First of all, Fedora should focus on solving mysterious kernel bugs like this or this which cause a machine to completely freeze unexpectedly, without an apparent and real reason, and this a few times a day.

Jul 11 12:31:22 192.168.1.4 dbus-daemon[1672]: [session uid=1000 pid=1672] Successfully activated service 'org.gnome.evince.Daemon'
Jul 11 12:31:22 192.168.1.4 systemd[1585]: Started Evince document viewer.
Jul 11 12:34:20 192.168.1.4 kernel: perf: interrupt took too long (6434 > 6407), lowering kernel.perf_event_max_sample_rate to 31000
Jul 11 13:01:01 192.168.1.4 CROND[4523]: (root) CMD (run-parts /etc/cron.hourly)
Jul 11 13:01:01 192.168.1.4 run-parts[4526]: (/etc/cron.hourly) starting 0anacron
Jul 11 13:01:01 192.168.1.4 run-parts[4532]: (/etc/cron.hourly) finished 0anacron
Jul 11 13:09:59 192.168.1.4 /usr/libexec/gdm-x-session[1605]: (EE) client bug: timer event3 debounce: offset negative (-192ms)
Jul 11 13:10:03 192.168.1.4 /usr/libexec/gdm-x-session[1605]: (EE) client bug: timer event3 debounce: offset negative (-304ms)
Jul 11 13:10:03 192.168.1.4 /usr/libexec/gdm-x-session[1605]: (EE) client bug: timer event3 debounce short: offset negative (-318ms)
Jul 11 13:16:50 192.168.1.4 kernel: ------------[ cut here ]------------
Jul 11 13:16:50 192.168.1.4 kernel: kernel BUG at mm/page_alloc.c:2019!
-- Reboot --
Jul 11 13:19:12 localhost.localdomain kernel: microcode: microcode updated early to revision 0xc2, date = 2017-11-16
Jul 11 13:19:12 localhost.localdomain kernel: Linux version 4.17.4-200.fc28.x86_64 ([email protected]) (gcc version 8.1.1 20180502 (Red Hat 8.1.1-1) (GCC)) #1 SMP Tue Jul 3 14:06:39 UTC 2018

Booting an older kernel (4.16.15) seems to solve the problem.

[angrypie]

Don't reply if you don't understand the issue.

[Weasel]

The most interesting festure of clear linux is the function multiversioning, with this the whole system runs like it has been compiled with -march=native

But... dude... that adds so much cruft and stuff you don't use on your particular situation, wrecking performance! Slim it down to exactly what you need then you get bare-bones and warp speed performance.

Maybe I need to throw in some "security" buzzwords as well, if performance isn't gonna cut it, to make it seem more important, like I have a point.

/sarcasm

[RahulSundaram]

First of all, Fedora should focus on solving mysterious kernel bugs like this or this which cause a machine to completely freeze unexpectedly, without an apparent and real reason, and this a few times a day.

Fedora is not a centrally managed project like that. The volunteer building this alternative kernel couldn't be redirected to work on some bugs you want them to prioritize. It is the not the same level of skills. It is unlikely either of these bugs are Fedora specific anyway

[Vistaus]

This has been tried multiple times for multiple distros. Last time Michael tried it in January for Ubuntu, all he managed was a decreased boot time with the Clear Linux optimized kernel. Nearly all other tests he ran were the same or showed only miniscule gains compared to the stock Ubuntu kernel.

I'm pretty sure that the reason ClearLinux wins the benchmarking tests is not because of optimizations, but because the distro, not being a real, robust distro, has very low overhead. It's built from the ground up for one thing - to win benchmarking tests.

I do notice the difference on Solus (which also uses many of Clear's optimizations).

[andyprough]

I do notice the difference on Solus (which also uses many of Clear's optimizations).

Solus also has advanced package management and a customized desktop. I'll bet you are getting a variety of performance improvements and lower overhead in general.

[Thaodan]

To get clear linux performance in most of phoronix tests its enough to export CFLAGS and CXXFLAGS

theres a patch from gravsky to set appropriat marches for the platforms.
funny but this isnt used in that case.


i use linux pf wich has the patch applied and build versions for skylake , silvermont and broadwell.