9 Aug
2012
9 Aug
'12
6:40 p.m.
On Thu, Aug 09, 2012 at 12:46:38PM +0200, Robin Gareus wrote:
On 08/07/2012 10:47 PM, Ralf Mardorf wrote:
I totally agree. I've been compiling my own
kernels since 2.0.x, both for
fun and because, especially at the time, size and speed mattered.
However, as the kernel grew in intricacy, especially so in the last
couple years, the trade-off between time investment and
performance/reliability seems to have swung on the debit side of the
pocketbook. This especially became clear to me when I worked on
compiling a kernel for my new netbook in hopes of improving battery-life
only to see my custom kernels consistently drain *more* power than the
Debian stock kernel. In the end, the time spent configuring the system
itself was far more effective, and I have a nagging suspicion that it
may prove to be the same with RT performance.
Cheers,
S.M.
On Tue, 2012-08-07 at 13:39 -0700, Ken Restivo
wrote:
The compile time is not even 1% of what it takes.
Tweaking the config, knowing and learning about the effects of all the
kernel parameters, how they interact and affect the system is a
never-ending task. It is very easy to screw it up, too.
IMHO it is much wiser to spend the time to contribute to a community effort.
Sure, you can tweak the kernel to your personal system and preferences
when you roll your own. You can learn a lot by doing that and it can be
fun, too.
At 64studio we've learned that the actual performance difference
regarding options that are relevant only to one system vs options that
are suitable for a distributions are minimal. Debian continues that
effort. ..and you get updates. Interesting, well I've already exceeded my
time budget for "while I've
got this thing on, I wonder if there's a 64-bit kernel for it".
You should build your own kernel-rt. In the end this might be less time
consuming. On my machine it takes around 90 minutes with an Athlon
2.1GHz dual-core, 4GB RAM, CONCURRENCY_LEVEL=2.