Pay For Faster Linux Kernel Performance? There's Patches For That

[LLStarks]

This reminds me of Anthrax Kernels

Sell binaries and publish incomplete code while keeping the compilation techniques under wraps

[Uqbar]

Linus' thoughts?

I wish I had some Linus' thoughts about this whole story.
My personal one.
I mean, technically speaking. I am not really interested in the political/economic/whateverelse point of view.
I wish they dared to push those changes to the kernel dev list for review with two way communication to provide explainations.
If they don't, then chances there are it's all a load of bu11sh!t.
In the end, we are dirty techies, don't we?

[marvin42]

compiler flags

you can easily compile with your own flags by using e.g. make bzImage CFLAGS_KERNEL="-march=native". As this mostly tunes vector operations (SSE, AVX) it shouldn't change much because kernel doesn't use them. Also this may uncover some kernel or compiler bug resulting in an unstable system and is also not supported. Maybe phoronix may publish results of a test run, but I doubt it will change much.

[geearf]

The website seems pretty clear, I am not sure about all these questions:

Kernele zostały zbudowane w oparciu o źr?dła Vanilla ("czysty" kernel) kernela 3.14.x. W tym przypadku podobnie, jak już to uczyniłem wcześniej, zastosowałem auto-grupowanie (Automatic process group scheduling). Poniższe kernele używają defaultowoschedulera I/O CFQ.

which translates to:

kernels are built based on the source of Vanilla ("clean" sources) kernel. Referring to their configuration, I used the auto-grouping (Automatic scheduling process group). As defaultowego scheduler processes / CPU load use CFS, while for process input / output [drives] scheduler I / O CFQ.

Another interesting excerpt:

However, due to the use of my very deep level of compiler optimization, built on the basis of their kernels will be 100% match I shared kernel packages in terms of functionality, but do not match them in terms of performance and responsiveness on the architecture for which they were intended . First of all this is true kernel of the line i7 and Brazos, which are the deepest optimizations in the process of building them at compile time.

Due to the fact that the GPL 2 license under which the kernels are distributed, it does not require me to disclose the process (rules and commands) compilation (talking only about the scripts implementing this process - and these reveal) finished packages and the fact that setting These are kind of my trade secrets, for the moment will not be disclosed.

I join the guy that linked to funloop so far.
Of course that is probably because I do not have a "very deep level of compiler optimization" knowledge.

[fguerraz]

Due to the fact that the GPL 2 license under which the kernels are distributed, it does not require me to disclose the process (rules and commands) compilation (talking only about the scripts implementing this process - and these reveal) finished packages and the fact that setting These are kind of my trade secrets, for the moment will not be disclosed.

That's bollocks, from the GPL v2 licence:

For an executable work, complete source code means all the source code for all modules it contains, plus any associated interface definition files, plus the scripts used to control compilation and installation of the executable.

[ext73]

Hi, I hope that the following will explain some queries:

First of all - to the Polish readers - please read carefully what is written on our website - www.netext73.pl.

Secondly - for the rest of readers:
- My kernels are based on vanilla sources ? it takes just a few seconds for experts in this subject to determine this.
- I release sources of my kernels modified by below elements and I do not conduct git - for now I do not have enough of time to do this.
- I use only the following modifications: the clock - HZ, ie introducing additional clock HZ = 500 - in my opinion this is the most appropriate solution and I modify TSC [tsc.c] that means I?ve changed any notification from the error to the info. This resultes in lack of annoying requests at the start of the machine. In addition I?ve added a aufs3 patch set.
- Performance and responsivenes results in the following: the kernel configuration and the application of appropriate kernel compilation + connection of the scripts that I created APM (energy management and efficiency of the machine) + the appropriate changes in the grub and installation the latest firmware and microcode for the proper CPU. All the operations motioned above are implemented by our program NeteXt'73. Such an environment is complete set for my kernels.
- I always write what and which version of the compiler I?ve used (you do not even make the effort beacouse I write in two languages , vide for example:

http://www.netext73.pl/2014/12/kernele-premium-v326-linux-3176kernels.html

I usually give a few lines of kernels:
1. Available free of charge - lines K8 [Intel and AMD - from the Pentium 4 and AMD K8 from (supports SSE3)] and kernels atom line [Intel atom].
2. For a normal fee [Premium lines] - i7 lines [Sandy/Ivy Bridge], i7-pro [Haswell], Brazos [AMD from A10.5 - Phenom II, Athlon II and Llano, Brazos, Bulldozer], Brazos-pro [AMD Piledriver, Steamroller, for example: Vishera, Kaverii, Kabini].

- I also give the packages with drivers for Nvidia and AMD [Catalyst] - the latter are being optimized and adapted to the latest kernel, Nvidia rarely - only if there is a need to adapt to the new kernels if they lack such support.
- Mentioned APM ? to see my system power management and performance please look at the following results:

http://www.netext73.pl/2014/03/kolejny-may-rektord.html

- Our NeteXt'73 - contains a number of modules that support the work of the system, for example: Sysv and upstart services management (in 1 window), grub settings, microcode/firmware, instalation of the newest application/adding external repos, etc. We support the Polish and English languages and approximately 80-90% in French language.
- In conclusion - I DO NOT turn on additional patches outside that are mentioned above.
- and relating to the GPL 2.0:

Thus, the license states:
The source code for a work means the preferred form of the work for making modifications to it. For an executable work, complete source code Means all the source code for all modules it contains, plus any associated interface definition files, plus the scripts used to control compilation and installation of the executable.

As I wrote [as a lawyer] I literally read the contents of the license, as a kind of contract between the parties. NO interpretation from my side. It is assumed that if the creator of the licence would mean: the rule compilation, which refers directly from the command line to the compiler and are not recorded in the file that controls this process = script. In this case the licence creator would have written:
?and commands/rules used to control compilation and installation? but he didn?t and wrote the following:
?plus the scripts used to control compilation and installation of the executable?.
I therefore assume that I act in full compliance with the contents of the license, as I publish the source code of any changes that are used in their kernels, including the configuration files associated with the compilation of the kernel.

Regards

e X t 7 3

[Kano]

Some of your changes are absolutely pointless because:

a) Ubuntu kernels have got already aufs (for Kanotix we compile the sources with static ahci, tmpfs with support for xattr and sometimes hotfixes or updated to latest dot kernel updates)
b) I tested hz changes for responsiveness years before you, they are basically useless, even hardcore gamers could not notice any difference
c) I don't get what you need to patch on Grub, i do not need Grub at all on UEFI systems, i can replace it with Gummiboot or boot directly with the kernel (thats the reason for the static ahci) even without initrd (best with gpt to use root=PARTUUID=xxx) - thats pure speed with disabled CSM and fastboot (in case you only want to boot from 1 ssd)
d) Are you really sure that you mean APM? It was replaced by ACPI years ago.
e) Microcode updates for CPUs are in the firmware usually, there you can just update that. Maybe for outdated boards it is usefull if use with a newer CPU than usually designed, but then you could install the needed package in userspace to update it. But iucode-tool/intel-microcode is updated without you as well.
f) You are not the only one who updates nvidia/fglrx drivers to work with newer kernels. I do that for Kanotix (or if somebody with plain Debian uses my repository) as well. I would not say they are "optimizied" as this is impossible. You can just add the same available patches as I do from any other distro/forum entry, whichever was first to support it.
g) Your kernels are outdated, why don't you use 3.18.x?
h) You must build your kernels in a very stupid way if you do not package the /boot/config-$KVER files, i would absoltely never use it, you can not even recompile it with extra patches. Absolutely useless that way.

[Guest]

This discussion starts to be interesting. "I'm building a distro. I'm building kernels. My di** is longer. No, my di** is longer!" Can you choose your female representant and start fight?

http://fc01.deviantart.net/fs34/i/2008/309/9/9/catfight_by_PaddysDemon.jpg

[ext73]

Some of your changes are absolutely pointless because:

a) Ubuntu kernels have got already aufs (for Kanotix we compile the sources with static ahci, tmpfs with support for xattr and sometimes hotfixes or updated to latest dot kernel updates)
b) I tested hz changes for responsiveness years before you, they are basically useless, even hardcore gamers could not notice any difference
c) I don't get what you need to patch on Grub, i do not need Grub at all on UEFI systems, i can replace it with Gummiboot or boot directly with the kernel (thats the reason for the static ahci) even without initrd (best with gpt to use root=PARTUUID=xxx) - thats pure speed with disabled CSM and fastboot (in case you only want to boot from 1 ssd)
d) Are you really sure that you mean APM? It was replaced by ACPI years ago.
e) Microcode updates for CPUs are in the firmware usually, there you can just update that. Maybe for outdated boards it is usefull if use with a newer CPU than usually designed, but then you could install the needed package in userspace to update it. But iucode-tool/intel-microcode is updated without you as well.
f) You are not the only one who updates nvidia/fglrx drivers to work with newer kernels. I do that for Kanotix (or if somebody with plain Debian uses my repository) as well. I would not say they are "optimizied" as this is impossible. You can just add the same available patches as I do from any other distro/forum entry, whichever was first to support it.
g) Your kernels are outdated, why don't you use 3.18.x?
h) You must build your kernels in a very stupid way if you do not package the /boot/config-$KVER files, i would absoltely never use it, you can not even recompile it with extra patches. Absolutely useless that way.

Hi

I will try to respond to your post in the evening. At this point, I have very little time, so briefly:

- APM is my set of scripts to manage the power and performance of the machine while working AC/BATT - a few of the elements of the configuration:

APM Menu - NeteXt'73

- If the changes introduced by me would be pointless ... what miracle such test results under PTS ?

Regards

Tomek

[Kano]

I don't think you do more than powertop suggests. If you need to write a new gui for that, have fun.