Fast Kernel Headers Work Restarted For Linux To Ultimately Speed Up Build Times

[stiiixy]

None, zero, zilch, squat, jack shit, that's how much.

Want to know why?

Because I am not a sucker, that's why!

Linus makes 1.7 million dollars a year, the Linux Foundation makes 177 million a year, Alma Linux makes at least a million a year, Rocky Linux raised 26 million dollars in startup capital and makes millions a year and you think i am going to write a single line of code for free?

Fuck no!

When one of these projects pays me a good salary to write code or when they sign a legally binding contract to share their revenue with me, then I will contribute code.

Most GPL contributors are straight up suckers. The GPL is about screwing over your fellow man, get them to work for free, take their code, make a shit load of money from it and them complain if someone else want to benefit from code you never wrote.

The GPL is one of the biggest scams every perpetrated and if someone wants me to contribute anything they need to make it worth my time.

How much did it cost you to write this up?

[dEnigma]

Unless this work is going to make these build times drop from 90 minutes to 20 minutes / four hours to one hour / eight hours to two hours-ish, they are of little relevance.

Well, according to the original Phoronix article about this effort:

According to Ingo's figures, there could be as many as 78% more kernel builds per hour with the "Fast Kernel Headers" enabled kernel than the current stock kernel. A 50~80% improvement in the absolute kernel build performance on supported architectures is possible.

Not sure why everything that doesn't at least quarter the compile times would be of little relevance.

[Sonadow]

Just had a look at the mailing list and the megapatch breaks building in so many ways it's not even funny.

[X_m7]

Improvements to kernel build times would be great, as someone who has to patch and recompile my own kernel to not get power throttled for no good reason on my HP Omen 16 laptop, at least until upstream accepts the patches and they make it down to a stable kernel.

I just did a kernel build of the Arch Linux 6.7.2.arch1-2 kernel plus the patches using the `time makepkg -s` command on that Omen 16 laptop (with a Ryzen 7 6800H, 8C/16T, but frequency capped to 3.75GHz to avoid overheating) and with the build set up to use 12 threads for a bit of leeway on other things, and the whole process (including verifying the kernel source tar with GPG, packaging etc) took almost exactly 31 minutes.

Edit: I also stripped the documentation building part out of the PKGBUILD.

[sophisticles]

And for that to happen, a stable unchanging kernel ABI is required so that a driver package for kernel version X can still be installed on kernel version X,Y or even kernel version X+1, and that hardware vendors don't have to waste time building modules against every single damn version of the kernel.

But we all know why that has never happened for Linux.

Also, how is it possible to build a allmodconfig in only 133 minutes when it takes a Supermicro workstation with two E5 2690-v2 processors 60 - 90 minutes to build my kconfig, even when the whole build is done entirely on a 80GB ramdisk. Any secret sauce?

The "secret sauce" is that Ice Lakes have AVX-512 and they are reasonably potent.

According to Michael's benchmarks, the 8700G can do allmodconfig in 19 minutes 12 seconds:

https://www.phoronix.com/review/amd-ryzen7-8700g-linux

A default build takes only a minute and a half on the 8700G.

This is why I said this work is such a waste of time.

[quaz0r]

rewrite lunix in c++ with modules

[rmfx]

I think it's a serious and sad waste of resources that fast kernel header were achieved, then nothing was done to merge it...

I don't know why this dev is starting over instead of improving upon, but anyway, that's still good to see.

[smitty3268]

Also, how is it possible to build a allmodconfig in only 133 minutes when it takes a Supermicro workstation with two E5 2690-v2 processors 60 - 90 minutes to build my kconfig, even when the whole build is done entirely on a 80GB ramdisk. Any secret sauce?

That would be 16 cores of a Sandy Bridge architecture cpu, which was launched in 2011.

CPUs have gotten quite a bit faster since then.

[yoshi314]

I must be missing something, a 7700X, 8C/16T, is capable of building the kernel in under 79 seconds, that's not fast enough?

Linus's system builds it in 22 seconds, how much faster do they think they can go?

just because it's fast for one build doesn't mean it's fast for a test array of builds. or low-spec hardware.

i would assume there are big gains to be made in CI/CD.

[fallingcats]

A stock distribution config built on a Supermicro workstation with two E5 2690-v2 processors entirely on a 80GB ramdisk takes almost 30 minutes.

That same stock distribution config built on an i7 Skylake notebook takes almost two hours.

That same stock distribution config built on a Gemini Lake notebook takes almost four hours.

My kconfig requires 60 - 90 minutes on the Supermicro workstation, four hours-ish on the Skylake notebook, and an entire working day on the Gemini Lake.

Unless this work is going to make these build times drop from 90 minutes to 20 minutes / four hours to one hour / eight hours to two hours-ish, they are of little relevance.

Because your workstation has 20 cores from >10 years ago running at 3GHz with DDR3. Meanwhile IPC has at least doubled, memory bandwidth quadrupled (a modern SSD is within an order of magnitude from your RAM), and efficiency gains from a flatter memory layout not using two processors are real.