You are not logged in.
Pages: 1
Am I a pro now?
Actually, I'm wondering if there should be any performance difference now. I didn't patch it at all. I just took the default kernel, adjusted the timer frequency and took out a ton of drivers and modules that I don't need. The kernel is a lot smaller in size.
Did I do anything for performance or do you have to use some combo of patches to change performance?
I'm glad I did it anyway, it was fun.
Offline
Most likely no performance boost, no.
You may have a faster boot time though, and more Linux experience -- that and fun are the two real good reasons to custom-compile a kernel when stock works acceptably, IMO. And street-cred!
Last edited by Ranguvar (2009-09-28 01:40:38)
Offline
He's right c:
Offline
If you only take out drivers you don't need, you won't really gain any performance, because these modules wouldn't have been loaded anyways. What makes your kernel faster are things like the timer frequency, take out things like "generic x86 support" etc, stuff you and your machine don't need. The kernel in the repos needs to fit any machine, so such things can't stay out, but you know that you don't need it, so leave it out and gain some speed.
I have a heavily modified zen-kernel, I really threw everything out what I don't need. And what do I have to say? It isn't as much more performant as I had expected. I still keep using it, because it IS a little bit faster and there's one thing a custom kernel has: more security! If there's a security lack in some sections of the kernel you haven't even compiled in, you can't get infected. A kernel working on every machine will rather have this compiled in.
Offline
It's a myth. Even compiling for your architecture does not yield better performance, except for Core 2 Duo. Stuff like sse3/4 will be exploited by apps.
If you want a quicker boot, compile in the modules (don't build as module, include it) and skip udev. Honestly, these days it doesn't matter. Before i686, it did.
I need real, proper pen and paper for this.
Offline
Agreed I really would bother compiling your own kernel. But kudos to you anyway! Back in the day on my 468 dx2 I seen a big improvement compiling my own kernel... But seeing that my machice is up 24 x 7 faster boot times don't matter to me.
Offline
It's a myth. Even compiling for your architecture does not yield better performance, except for Core 2 Duo. Stuff like sse3/4 will be exploited by apps.
If you want a quicker boot, compile in the modules (don't build as module, include it) and skip udev. Honestly, these days it doesn't matter. Before i686, it did.
I would disagree that it is not a myth. I compiled openoffice and I saw a huge performance increase. Also look at ubuntu. Many people believe certain apps are slower on ubuntu than say on arch. I believe Ubuntu compiles their pkgs for i486. However, in some cases you will not see any noticeable performance gain. The way I understand the kernel build process, is that even if you set CFLAGS they get unset.
@OP if want to get your hands dirty I would try patching the kernel with bfs
Offline
The way I understand the kernel build process, is that even if you set CFLAGS they get unset.
If you want to give this a try, use the zen kernel, there's a section where you can put your custom cflags in
Offline
I would disagree that it is not a myth. I compiled openoffice and I saw a huge performance increase. Also look at ubuntu. Many people believe certain apps are slower on ubuntu than say on arch. I believe Ubuntu compiles their pkgs for i486. However, in some cases you will not see any noticeable performance gain. The way I understand the kernel build process, is that even if you set CFLAGS they get unset.
@OP if want to get your hands dirty I would try patching the kernel with bfs
Wait, you would disagree that it is not a myth? Then you would agree that it is a myth!
Correct, but I wasn't talking about apps - just the kernel. And moreover, it's a select few which really "improve". Other magnitudes of "improvement" exist, but are not significant for any machine >= 800MHz and 256MB.
But if on a Core 2 machine (or an AMD equivalent), you will benefit by optimising for the arch.
I need real, proper pen and paper for this.
Offline
Pages: 1