perlun.eu.org · About the blog · Programming · Select language / Välj språk

Linux kernel compilation as a benchmark tool

Back in the days, compiling the Linux kernel was a rough (but still remotely useful) measure of the performance of a machine. How long does it take to compile a modern Linux kernel on a modern piece of hardware? Let's find out!

I started off by downloading kernel 4.17.4 from https://www.kernel.org and unpacking it:

$ wget https://cdn.kernel.org/pub/linux/kernel/v4.x/linux-4.17.4.tar.xz
$ tar xvf ~/Downloads/linux-4.17.4.tar.xz

Then I was looking for my kernel config in /proc/config.gz or similar, but apparently it wasn't available there. /boot/config-4.16.0-2-amd64 had it though.

I tried running make oldconfig to use this config (and add potentially new configurations from 4.17.4 to it):

$ make oldconfig
  HOSTCC  scripts/basic/fixdep
  HOSTCC  scripts/kconfig/conf.o
  YACC    scripts/kconfig/zconf.tab.c
  LEX     scripts/kconfig/zconf.lex.c
/bin/sh: 1: flex: not found
scripts/Makefile.lib:188: recipe for target `'scripts/kconfig/zconf.lex.c' failed
make[1]: *** [scripts/kconfig/zconf.lex.c] Error 127
Makefile:528: recipe for target 'oldconfig' failed
make: *** [oldconfig] Error 2

sudo apt-get install flex to the rescue. I re-ran make oldconfig and answered "No" to most new things added in 4.17.4. (I just went with the suggested defaults.) And there were quite a lot of them actually, for being a minor release!

Then I attempted to start the kernel compilation. Since my CPU is a 6-core Ryzen with hyperthreading, I used make -j12 to allow up to 12 parallel gcc jobs - it failed first because of yet another missing dependency:

$ time make -j12
scripts/kconfig/conf  --syncconfig Kconfig
Makefile:970: *** "Cannot generate ORC metadata for CONFIG_UNWINDER_ORC=y, please install libelf-dev, libelf-devel or elfutils-libelf-devel".  Stop.

real    0m1.625s
user    0m1.039s
sys 0m0.852s

Once that had been resolved, it said bc: not found on the next attempt. Then, I got an error about certificates:

$ time make -j12
[...]
make[1]: *** No rule to make target 'debian/certs/test-signing-certs.pem', needed by 'certs/x509_certificate_list'.  Stop.
Makefile:1063: recipe for target 'certs' failed
make: *** [certs] Error 2

This was because my .config referred to an non-existing file; it exists inside the Debian-patched kernel tree, but my kernel source was a pristine, upstream tarball with no Debian-added files present.

I edited my .config, removing CONFIG_SYSTEM_TRUSTED_KEYRING and CONFIG_SYSTEM_TRUSTED_KEYS and re-ran make oldconfig and make clean.

After that, it finally worked and I could sit back and relax.

$ time make -j12
  LD [M]  sound/usb/snd-usbmidi-lib.ko
  LD [M]  sound/usb/usx2y/snd-usb-us122l.ko
  LD [M]  sound/usb/usx2y/snd-usb-usx2y.ko
  LD [M]  sound/x86/snd-hdmi-lpe-audio.ko
  LD [M]  virt/lib/irqbypass.ko

real    16m14.810s
user    168m15.336s
sys 15m37.348s

16 minutes - IIRC, I think it used to take about 6 minutes about 20 years ago, so the figure right now feels a bit high. The machine used back then was obviously not as fast as this one (which has a super-nice M2 PCIe SSD drive), but OTOH the sheer size of the kernel tree has likely grown quite significantly since then.

Just for the sake of it, I downloaded a kernel from "that era", namely 2.4.10 - released on September 2001.

$ ls -lh linux*
-rw-r--r-- 1 per per 22M Jul  8 20:48 linux-2.4.10.tar.bz2
-rw-r--r-- 1 per per 98M Jul  8 07:33 linux-4.17.4.tar.xz

So the new kernel tree is approximately four times as large as the old one, if we assume that bzip2 and xz are equally efficient in compression (which is likely not entirely true; my guess is that xz produces a somewhat leaner file).


Looking closer at the kernel tree where I did the compilation, the total number of .o files (i.e. ELF object files) generated during this compilation was 15841, and the total compilation time was 16 * 60 + 14 = 974 seconds. That gives me a total of about 16 files being compiled per second on average during this compilation cycle, which is pretty impressive. So slighly below 1s per file on average, given the "12 parallel jobs" setup - this seems moor than reasonable to me.

So the bottom line: yes, the machines are indeed much, much faster nowadays, but the Linux kernel has also grown significantly in size since then, making the compilation still take a bit of time - even more than before. On the other hand, most of us very seldom need to compile a custom kernel these days so maybe it doesn't really matter so much.