Does compiling the kernel makes the system run significantly faster?

smitty · 2008-02-05 04:29:09

It used to be when you compiled your kernel (at least ten years ago), you would get a significant speed boost. However, with the speed of machine nowadays, I wonder how much, if any. Does compiling the kernel have a significant increase in daily performance or otherwise? If so, what portion of the system does it makes faster than before? In short, Is it worth it? I appreciate any comments.

kishd · 2008-02-05 04:39:17

Tried compiling the kernel with various tweaks to increase system performance and really did not find any difference in daily use. I now only re compile the kernel for one of my machines to enable PAE so that I can use all the memory on that machine.

Allan · 2008-02-05 05:48:50

I was once bored and tried it out. Short answer was I didn't notice any significant speed increase. There is really not that many extra optimizations added when you go from i686 to current model chips.

iphitus · 2008-02-05 07:15:42

No. You'll get some ricers that claim it does, though it's typically placebo.

edit: I should add, that once in a while they make a change between kernel versions that makes things run a bit better -- but you'd hear about it if it was significant. There _was_ also CK, which though it didn't make things "significantly faster", it did improve responsiveness.

Last edited by iphitus (2008-02-05 08:38:40)

cry0x · 2008-02-05 07:38:37

It's not that you compiled your kernel, it's that you compiled your kernel without the extra features that you have no need for. This may or may not give you a perceivable performance increase, but theoretically some things should operate faster.

The only real speed increases I have seen regarding the kernel is when I've changed to a newer version with some optimizations. Changing from 2.6.23 to 2.6.24 shaved a few seconds off my boot time, and made my suspend/resume INCREDIBLY fast. At first I thought suspend was broken because it would sleep and wake so quickly... but no, it's just fast as hell.

robertp · 2008-02-05 08:16:19

Short answer NO. Compiling kernel makes sense when:

1. doing kernel development
2. needing very obscure feature which is not already available as module and hasn't been compiled into kernel

jbromley · 2008-02-05 08:19:50

No. For a modern 2.6.x kernel you probably won't notice the difference, even if you only configure what you need. Since a lot of the kernels is modules, you may have modules for hardware you don't have or features you don't use, but the worst they'll do is take up hard drive space and perhaps slow module loading a tiny bit. Of course if you use different compiler optimizations you may see a bit of difference.

If you want better responsiveness from the kernel (an improvement in responsiveness is often perceived as speed) you may try patching your kernel. This was what the ck kernel patches were all about in their day -- tuning the kernel to get better desktop/workstation responsiveness. Nowadays I'm not sure what a good patchset is.

If you've never done it before, build one working kernel and call it a day. It's a good thing to know how to do, but any real or perceived speed increase is probably not worth the time you spend building the kernel.

Regards.

sozo · 2008-02-05 09:26:37

Tried it a few weeks ago and measured with super_pi (5 runs):

stock kernel vs system specific compiled kernel gave a speed increase of 0.2 secs (1.5%), so I just use the stock kernel

broch · 2008-02-05 14:38:57

in my case customized kernel :
sped up boot ~ 10s (measured with bootchart) when comparing to default kernel
added a lot of security (grsec patch or RSBAC)
facilitated management (Arch has external modules that can be added to kernel making some updates possible e.g. ipw3945 working with kernel 2.6.24 which is not possible with Arch modules as first iwl driver was really crappy)
better memory management (-ck/post -ck patches) generating better responsiveness under heavy load

did not improve general system speed

also, once in a while there are boot problems with updated kernels. Never happened to me as I always have one trusty kernel that boots.

for references regarding gains from different patches, check gentoo site and search for benchmarking of different patches vs default vanilla.

kensai · 2008-02-05 14:49:47

In reality it does, but does not go that noticed, only if you are a gamer. Take for example Enemy Territory: Quake Wars, to play that game you need your kernel to be compiled with Kernel Low-Latency Desktop option and 1000Hz. Arch doesn't provide this optimizations by default, so if you want to get the best out of your games you have to compile your kernel. But Linux kernel compilation is too tedious and I hate doing it, although I just go and check this and this and uncheck that and that and compile in less than 20 minutes. On the other hand FreeBSD is so organized it is great and fun to compile the kernel.

Short Answer: Yes, depending on what you use your system for.

jacko · 2008-02-05 14:55:10

I compile the kernel just for something to do. See how much stuff I can remove and still make my computer functional.

patroclo7 · 2008-02-05 15:40:14

Definitely no advantage if you SIMPLY recompile it (without configuring, cleaning and personalizing it). If you configure, clean and personalize, then: faster boot; some minor, minor optimizations; many possible breakages; a very instructive and pleasant experience, if you spend a bit of time reading help files and kernel docs.

lilsirecho · 2008-02-05 16:40:52

The fastest system utilizes ram for operations.

In a nutshell, eliminate the slow items in the system by reducing the system load for one approach.

Using a "live" system such as Faunos (500 packages) with a compact flash to IDE adapter provides a 45 second boot time. All programs are executed in ram.

Use of a second compact flash device with IDE adapter allows the user to archive all packages desired and provide a pacman install interface script for same. These installed programs reside in ram and execute in ram. Upon reboot and no save they are not saved. Reboot can in this instance be a reset of the computer which is instantaneous.

Speedy!!!! The ram in use is basic booted ram plus any installed program. The flash archive is mainly r/o. It can hold GB of video, mplayer,vlc,avidemux and all deps.

Fastest thing I know for arch linux (opens as arch linux when root is selected).

Cannot be ruined by outside activistas!!

broch · 2008-02-05 21:02:58

Using a "live" system such as Faunos (500 packages) with a compact flash to IDE adapter provides a 45 second boot time. All programs are executed in ram.

well that is how linux or bsd work.

45s for boot time is not what you are referring as fast?

not to mention that the original question was about kernel.

Dusty · 2008-02-05 23:04:42

I strongly doubt you're going to get a speed increase that, over the life of the kernel, is going to save you more time than you spent compiling it.

lilsirecho · 2008-02-05 23:27:50

Broch:

The title of this thread doesn't limit responses to just redoing the kernel since it says "system run significantly faster".

The kernel is not the system...the system runs from HDD and ram.

To speed up that process, run everything possible in ram for system speed up.

I think for 500 packages loading from scratch to desktop in 45 seconds is fast. And running after that in ram is super fast system speed "significantly faster".

The speed I quote is possible now. With UDMA capability in the kernel, at least half of the install time is probable.

And that is for over 500 packages into KDE no less!

Using flash in IDE removes the latent search from the system and therefore must be faster.

Changing the kernel will indeed speed up the system...but only if it includes UDMA for the flash drives already available but not configured with the present kernels.

Using flash drives for ide cache repos will allow faster boot time, and reduce the load on the system. You don't normally run many,many programs at once so put them in cache with a script to permit fast loading .

That is "system running significantly faster" and all in ram, the fastest in the computer.

dhave · 2008-02-06 00:05:16

lilsirecho wrote:

Using a "live" system such as Faunos (500 packages) with a compact flash to IDE adapter provides a 45 second boot time. All programs are executed in ram.
...
Use of a second compact flash device with IDE adapter allows the user to archive all packages desired

Is using an IDE adapter essential, or can the same performance be achieved via USB?

Last edited by dhave (2008-02-06 00:07:48)

lilsirecho · 2008-02-06 01:04:20

Usb is limited to 15MB or so by overhead on read and write.

When running in IDE, the HDD seek time latency is eliminated , and, and, if, if, UDMA is initiated the speed can go to 45MB/sec in UDMA 5.

USB cannot come close to that speed.

broch · 2008-02-06 04:27:36

lilsirecho wrote:

Broch:
The title of this thread doesn't limit responses to just redoing the kernel since it says "system run significantly faster".
The kernel is not the system...the system runs from HDD and ram.
To speed up that process, run everything possible in ram for system speed up.
I think for 500 packages loading from scratch to desktop in 45 seconds is fast. And running after that in ram is super fast system speed "significantly faster".
The speed I quote is possible now. With UDMA capability in the kernel, at least half of the install time is probable.
And that is for over 500 packages into KDE no less!
Using flash in IDE removes the latent search from the system and therefore must be faster.
Changing the kernel will indeed speed up the system...but only if it includes UDMA for the flash drives already available but not configured with the present kernels.
Using flash drives for ide cache repos will allow faster boot time, and reduce the load on the system. You don't normally run many,many programs at once so put them in cache with a script to permit fast loading .
That is "system running significantly faster" and all in ram, the fastest in the computer.

no

Does compiling the kernel makes the system run significantly faster?

compiling kernel, not whatever
I did not suspect that this requires explanation?

but if you really want the noise, let's leave "faster?"
and answer is: yes/no
whatever that means.

or you are joking

anyway seems that results may vary.

Last edited by broch (2008-02-06 04:32:01)

Arch Linux

#1 2008-02-05 04:29:09

Does compiling the kernel makes the system run significantly faster?

#2 2008-02-05 04:39:17

Re: Does compiling the kernel makes the system run significantly faster?

#3 2008-02-05 05:48:50

Re: Does compiling the kernel makes the system run significantly faster?

#4 2008-02-05 07:15:42

Re: Does compiling the kernel makes the system run significantly faster?

#5 2008-02-05 07:38:37

Re: Does compiling the kernel makes the system run significantly faster?

#6 2008-02-05 08:16:19

Re: Does compiling the kernel makes the system run significantly faster?

#7 2008-02-05 08:19:50

Re: Does compiling the kernel makes the system run significantly faster?

#8 2008-02-05 09:26:37

Re: Does compiling the kernel makes the system run significantly faster?

#9 2008-02-05 14:38:57

Re: Does compiling the kernel makes the system run significantly faster?

#10 2008-02-05 14:49:47

Re: Does compiling the kernel makes the system run significantly faster?

#11 2008-02-05 14:55:10

Re: Does compiling the kernel makes the system run significantly faster?

#12 2008-02-05 15:40:14

Re: Does compiling the kernel makes the system run significantly faster?

#13 2008-02-05 16:40:52

Re: Does compiling the kernel makes the system run significantly faster?

#14 2008-02-05 21:02:58

Re: Does compiling the kernel makes the system run significantly faster?

#15 2008-02-05 23:04:42

Re: Does compiling the kernel makes the system run significantly faster?

#16 2008-02-05 23:27:50

Re: Does compiling the kernel makes the system run significantly faster?

#17 2008-02-06 00:05:16

Re: Does compiling the kernel makes the system run significantly faster?

#18 2008-02-06 01:04:20

Re: Does compiling the kernel makes the system run significantly faster?

#19 2008-02-06 04:27:36

Re: Does compiling the kernel makes the system run significantly faster?

Board footer