>>> "gregory pecheret" <[EMAIL PROTECTED]> 22-Jun-00 7:44:25
PM >>>
>Novice remarks about micro kernel performance:
>In a way, i think monolithic kernel always be faster than
>micro kernel, because micro kernel is one more layer (like
>a virtual machine) to applications.
>There are IPC between micro-kernels and servers, but
>also between servers and applications.
>But we can't have both performance and all micro kernel
>advantages!
But then if you can't break down the code into self contained chunks
then it's going to be more difficult to optimize and therefore might
be slower (esp. as OSes have to do more and more).
There are really 2 arguments I think:
1. let's make kernels more modular
2. let's make kernels more modular and have the modules in user
space
1 doesn't necessarily imply 2.
Many RTOS systems are apparently now moving towards modularity but
all in kernel space (or supervisor mode of the processor) because it's
easier to deal with a lot of services that way (I guess Linux modules
could be thought of in that way).
What this DOESN'T get you is crash protection: if one element crashes
likely as not the rest will too. You'll have to buy the whole
micro-kernel for that argument.
IMHO with Moore's law set to continue for the next 30 years we can
ease off on performance and try giving reliability a bit more
priority.
Nic
