In addition to measuring boot time and kernel size before and after re-compilation, what are some other useful metrics for determining if a custom-compiled kernel improves system performance? I know that performance increases will be slight, but I'd like to know if there are any differences before and after, anyway.

Edit: CPU is x86 32bit.

  • For boot time optimization just use systemd. – Ijaz Ahmad Khan Apr 22 '16 at 14:39

Assuming you're running x86_64 (amd64) architecture, don't expect a huge difference in performance. This architecture gave things a new baseline for processor features (as compared to 32-bit code possibly going back to i386).

Also, in the 32-bit world kernels and C libraries have already been compiled for different minimum architectures (i586, i686, ...).

Finally, things like games, compression libraries, crypto tools, 3D rendering, ... may already include multiple implementations of core functions that select the most efficient implementation for the current hardware.

But if you must: try to find a reliable standard benchmark suite that measures the performance characteristics that are important to you. Run it many times and look for consistent results.

Compile your own kernel then run it many times again.

If you see any difference, go back to the original kernel and see if the difference remains.

Then go back to your new kernel and see if the difference remains.

Many things can create small differences in measurements, and I don't expect you'll achieve any huge difference.


Simple: run a representative test while running old kernel and while running custom kernel and compare the results.

The key here is "representative": different workloads will have different performance requirements and may work better or worse on different kernel settings. Your custom-compiled kernel may improve performance for some apps while hurting others.

Either way, I would not expect a huge difference. It does not often make sense to use custom kernel, especially if you don't have specific goal in mind.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.