|
[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]
On Wed, Apr 09, 2003 at 07:55:43PM -0400, Toby DiPasquale wrote:
> On Wed, 2003-04-09 at 19:10, gabriel rosenkoetter wrote:
> > I happen to think that microkernel architectures will *never* be
> > worth the effort, that they're a bad case of putting the engineer
> > before the user to the vast detriment of the latter.
Linus has stated that all of the tricks that exist to get microkernels
to approach the performance of conventional macrokernels are just
tricks -- and are equally valid as macrokernel optimizations. So
if microkernels were easier to implement/extend, they might be more
useful to speed OS development; yet history has proven that this
distinction is irrelevant.
Also, the architectural differences themselves are rather moot.
Macrokernels today have loadable kernel modules and pluggable
architectures; is it really all *that* different from a generic
microkernel design?
The one last outstanding issue that remains to be resolved is in
"extreme plugability". OS X is built on top of Mach, and the Apple
engineers insist that it is one of the reasons why Darwin/OS X
"just works" when network connections magically appear and disappear,
when the computer suddenly goes to sleep or wakes up, etc. If that
level of performance/configurability can be done with a stock Linux
kernel in the next few years, then the distinction is truly academic. :-)
> Until you realize that Windows NT/2000/XP/2003 collectively is the
> world's most prevalent microkernel architecture and they indisputably
> put the user before everything else. (aside from themselves, that is)
Actually, NT started out as a microkernel architecture, back when Dave
Cutler was targeting the i860. Perhaps the first release or two on the
386 also retained some of that microkernelness. But as late as 1998 or
so, Microsoft stopped advertising the "microkernel" feature in the press
around NT, and Cutler started hedging the issue. I think NT4 is the
release that put the graphics subsystem into kernelspace for performance
and effectively negated any claim that NT was a true microkernel.
Z.
_________________________________________________________________________
Philadelphia Linux Users Group -- http://www.phillylinux.org
Announcements - http://lists.netisland.net/mailman/listinfo/plug-announce
General Discussion -- http://lists.netisland.net/mailman/listinfo/plug
|
|