It takes me just under 3 hours to build kernel packages with Pat's scripts. That's using my desktop which has the best specs. Is 3 hours good, bad or it depends?

Pat's scripts build all 5 packages (source, huge, generic, modules & headers). I only need the kernel-generic & the kernel-modules packages to build the initrd & run the custom kernels.

I started by following dive's helpful doc: Yet Another Kernel Compile Guide

I've setup a couple scripts that run from a KDE5 terminal or CLI tty (KDE5 logged out) at similar speeds:

1. setup the build directories & download the build files then
2. modify the config files, prepare the variables & run Pat's build-all-kernel.sh script.

I haven't built kernels for a long time so I'm a bit rusty on the best ways.

I have a new Ryzen 5900x system and I can compile just the kernel in 5 minutes. And when I did Pat's SlackBuild a while back I know it took less than 15 minutes for me.

2 members found this post helpful.

3 hours vs 15 minutes. That's a big difference! I need to find a better way.

I compiled new test kernels a couple of weeks ago on an 4-core i5 Skylake. Time was about 20 minutes.

1 members found this post helpful.

I've got an i5-5250U CPU @ 1.60GHz NUC with 16GB RAM and an 500GB SSD. I'll try a build on that box and see if it can turn a lower time.

EDIT:
I'll look into distributed compiles with icecream also. Based on this thread:[HOWTO] Set up icecream. I have 4 LAN machines with available cycles that I can use.

What about benchmarking your CPU againt the Ryzen 5900x ?

Ryzen 5900x
12 Cores, 24 Threads @3.7GHz, Zen3.
Release date: Q4 2020.

i5-5250U
2 Cores, 4 Threads @1.6 GHz, Broadwell.
Release date ≈ Q1 2015.

EDIT
No very clear message : your time seems correct to me regarding the difference between the machines.

That sounds promising, thank you for the heads up.

i7 Kaby Lake building Gentoo kernel with -j8
I remember my first ever kernel build on 486DX, kernel version was 2.0 and it took 14 hours ...

Edit: for those not familiar with time output, it took just under three minutes.

1 members found this post helpful.

Yikes! I feel the need, the need for speed!

Chuck56,

I cheated, this is very trimmed down custom kernel. Building a generic include-it-all would take much longer, I'm sure.

OK, I'm sold on using icecream to compile large builds like kernel packages.

Using icecream I combined my desktop and my Plex NUC. Pat's kernel build scripts went from 2 hours 56 minutes on only my desktop to 1 hour 50 minutes on the distributed compile pair. Since all my machines are low powered, low noise I'll add another PC or 2 to reduce the time even more.

Gotta love finding useful tools Pat makes available to Slackware users.

Guys, the kernel build times depend heavily on the kernel config and kernel version. For a meaningful comparison everybody needs to agree on the same parameters.

For reference, I've just built a 5.11.2 kernel essentially with full slackware generic kernel config in 4 minutes 33 seconds (Ryzen 5950X)

In the past I have used distcc to great effect, but distcc obviously works best when you have a number of machines of similar performance. If one machine is significantly faster than the others, you might as well build on the fast machine alone.

2 members found this post helpful.

Yeah, for actual 1-to-1 comparison you will need to compare the exact same compile on different computers.
I would argue that dot-versions of the kernels which are not too far apart (like maybe 3-5 releases or so) should normally give fairly similar compile times, given same/very similar configs, but obviously YMMV on that. Should give you an indication at least. Only the very same config and kernel version is straight up directly comparable.


I can't speak for distcc, but I have tested icecream a little bit, and it can speed up the compiles significantly - depending on what hardware you have available. As indicated, how well it works will depend on the differences between the computers you have available. Up to the point that some might actually slow down the overall compile. Some testing will figure that out for you. How helpful slow(er) machines are will also depend on the actual compile job. I would think that the bigger the compile job, the more helpful the slower nodes might be. Again, running tests to find out.

AFAIK Pat uses an icecream cluster for his compiles.

Thanks
--
KarlMag

If I had to guess you missed the make -jN flag (parallel build).

If he uses the -current trees' build scripts, in all the k/*.SlackBuild scripts you have this;
NUMJOBS=${NUMJOBS:-" -j$(expr $(nproc) + 1) "}

So, it should automagically figure out how many threads you have, add one and use that as you 'N'.


(Obviously - for a distributed (distcc/icecream) build you would have to set that number to something suitable yourself.)

Thanks
--
KarlMag

Here it takes about 5 minutes to build the kernel (5.11) with config-generic,
and about 35 minutes to build 'bzImage modules'.
I run 'make -j9', CPU is an AMD FX-8320E (eight cores).

2 members found this post helpful.