Re: Linux and Microkernels (was Re: your mail)

Alan Cox (alan@cymru.net)
Thu, 25 Jul 1996 09:52:19 +0100 (BST)


> >>>>> "AC" == Alan Cox <alan@lxorguk.ukuu.org.uk> writes:
> AC> A monlithic OS is faster than a microkernel based system.
>
> Statements about "all micro-kernels" based on experience with Mach will
> guarantee incorrect results. The speed argument is not true any more.

I'm working on benchmarks of three microkernels, including AmigaOS which
had a microkernel done right, and not Mach which (IMHO) is a mess, although
OSF mach is looking nice.

> what the interrupt latency of Linux on a 486DX2-66 is, but under QNX
> it is 4 microseconds. In fact, QNX's performance is so well tuned and
> fast that it is used heavily in industrial real-time processing
> applications.

QNX is specifically designed for hard real time. You'll notice that it
uses things like single page table sets to get performance. For a generic
application any "microkernel" will be slower because you either

a) Have message passing overhead of some form (which costs
instructions)
or
b) Its not really a microkernel at all, its just a well structured
monolithic kernel like Linux is now

Alan