Building Your Own Kernel: Still Necessary?
 

If I understand correctly (and I might not) most newer linux distros allow you to add modules to the linux kernel in almost a hot plug/plug-and-play approach.

Is there really any reason to build your own kernel? Other than just wanting to be an ubergeek or for a highly specialized project.

Does the added functionality or smaller footprint outweigh the fact that you lose the ability to automatically update your kernel when your distro releases an update?

Thoughts?

For me it depends. I always build my own kernel on older machines (I'm talking Celeron 400MHz w/ 256MB RAM, basically anything less than 1GHz and 256MB RAM) because the speed increase is noticeable. I'm sure you'd also do it if you were doing embedded devices (not that I have any experience).

Also occasionally one of the distros will ship with a particular "experimental" kernel feature disabled that I want to use.

But I would say 99% of the time I just stick with the stock kernel that comes with Debian.

Do I still build my own kernel? Yes!

Is it necessary? That depends on how you define "necessary".

I don't modify my hardware at a rate that would make "plugging
drivers on the fly" a matter of concern. I do like full control,
and performance, and a small memory foot-print.


Cheers,
Tink

Nothing to do with it. Package management handles that.

Short answer - yes.
Something you appear not to have considered; would you prefer to install code (kernel included) that no-one had bothered testing ???.

Using Debian stable, I'm not sure "no-one" had bothered testing it is quite accurate. :D But for an enterprise server, sure as shootin' you'd want to test everything in the test bed before taking it live.

Bingo.

That plus I like to remove all the stuff I don't need. The default kernels are loaded with drivers for hardware I've never even heard about. Why should I burden my system with them?

I compile my own, mostly because it reduces bugginess and increases performance and teaches you something. On my system for example, there are lots of bugs if I don't disable certain things, I know I could probably disable most of them by passing options to the kernel, but that's just laziness. Also, there are things which will increase performance, sometimes drastically. And, of course, it will teach you something, especially if you've never done it before. Or you can just take the blue pill like the majority. I wonder if you know where that will get you ...

Cipher wished he had taken the blue pill instead of the red.

I'd say that compiling the kernel isn't necessary, but it is fun. Most of the computers rolling out today are fine with stock kernels for desktop applications.

For the speed demons, compiling the kernel is a necessity. For those who want to learn, compiling is required. For those with old hardware or enterprise/business class hardware, compiling is a must.

I don't compile my own kernel as the standard Slackware generic kernel is fine for me. Yes, it does contain stuff that I don't need, but that's really not an issue for me.

I'm the opposite - the standard kernel doesn't include everything I need so I download a current kernel from kernel.org and compile away. Alien Bob's wiki is a good site for this.

It depends. For a workstation, I generally use the stock Slackware kernel. For servers, I strip them of everything I don't need.

depends, if i want a feature thats not present in the standard kernel shipped with whatever distro, then i'll do it.

example is tickless for 64bit. (because most distros' latest release was before the kernel came to suppor that)

that should be generally yes for a long use system regardless if it's a new system or an old system,.. a workstation or a server.. not temporary like live systems or the likes which would be impossible or impractical

Normally not - that way I went thru 3 boxes with the same drives and just had to plug them in.

I haven't done yet. But who knows what I might get up to in the future?