On Wed, Jan 16, 2008 at 02:16:40PM +0200, Lars Nood??n wrote:
| > ... (see ttys(5) for more info). But yeah, like Henning
| > said .. absolutely no need to build a new kernel.
|
| 1) Should anything be done to the GENERIC kernel's run time
| configuration then to improve performance, reduce system requirements or
| otherwise prevent it from making beer go flat?
Using config(8), you can simply enable or disable drivers if they
cause you problems.
| 2) Under what circumstances (generally) would one encounter a situation
| where it would strongly desirable to have a custom kernel?
I think the goal is "never". Under no circumstances *should* it be
'strongly desirable' to have a custom kernel. (this is my
interpretation, YMMV)
However.
If you're very low on resources (memory being the most important one
here, I think) or if you want to use some of the more experimental
drivers (NTFS comes to mind) or if you're testing / doing kernel
development, you may want to compile a non-GENERIC kernel.
I've been playing with bluetooth for a bit and have been testing ACPI
drivers (that were not enabled in GENERIC) for some time. These are
two cases where building a non-GENERIC kernel is called for. This sort
of stuff should not be used or run in a production environment (I
think). There's a reason why it's not enabled in GENERIC.
If you have a very constrained machine, memory-wise (like some
embedded piece of hardware) compiling your own kernel with all
non-required drivers removed may help. If stuff breaks, you get to
keep all pieces so make sure you know what you're doing in cases like
this. I'd say, double the RAM and be done with it.
Cheers,
Paul 'WEiRD' de Weerd
--
>++++++++[<++++++++++>-]<+++++++.>+++[<------>-]<.>+++[<+
+++++++++++>-]<.>++[<------------>-]<+.--------------.[-]
http://www.weirdnet.nl/
