According to Paul Clyne: While burning my CPU.
>
> Hi all,
>
> Got a question fro you that I havn't been able to find the answer to,
> possibly you can help. Appoligies in advance if the answer is located
> somewhere obvious or has been answerd before, Forgive me I come from a M$
> background..
>
>
> I understand the concept of a modular vs monolithic kernel.
>
> What is the speed tradeoff for each method ?.
>
> I have a P100 with 32M memory and 50M or so swap {which hardly seems to
> get used} but the systems _feels_ slow.
>
> Is a smaller Kernel (ie using modules) with the auto modle loader (what's
> it called again) running in the background going to operate faster than the
> same items made into a monlithic one ?. Or is the currently 'inactive'
> parts of a monolithic kernel going to suck more CPU cycles than the module
> loader would use.
>
> I realise that there are probably some other factors that effect the
> answer to this question (the number of modules for instance) but can I get
> a _general_ answer ?
>
Its not so much a question of speed directly in itself, or cpu usage thereof,
but a question of memory consumumed by a kernel having all options included
as opposed to modules. Its the actual kernel size which is at issue.
A good example would be a kernel built with all options (where possable) as
modules, the size of the kernel could be smaller than 300Kb, this means that
when the kernel is loaded at boottime it consumes very little memory, on the
other hand a kernel with built with options included within could excide the
640k barrier and at a certain size, would need to be built with bzImage, now
that one at boot time would consume a fantastic amount of memory.
So the basic princeple when makeing a kernel with modules is memory.
Keeping the actual kernel size to a minimum saves memory and therfor
increase performance, memory is "only" consumed when a particular module is
loaded automaticly by Kerneld and as soon as that module has not been used
for more than 60 seconds it will get automaticly unloaded along with any
dependant modules, ie something like the floppy driver, as soon as the
device is unmounted and unused for 60 seconds Kerneld will unload the
module(s) and the memory is then freed.
As a small experiment you could see for yourself what happens, make 2
kernels one with all options (where possable) as modules, the other builtin.
Install both kernels with lilo and reboot, after logging in use "free" to
check your memory, you will see very different results between the 2.
> thanks in advance..
>
>
> -------------------------------------------------------
> Paul Clyne aka: pacman / pac
> at work : [EMAIL PROTECTED]
> at play : [EMAIL PROTECTED]
>
> The future is in our hands. Which way to the future ?
> --------------------------------------------------------
>
--
Regards Richard.
[EMAIL PROTECTED]
