[EMAIL PROTECTED] wrote:
On Tue, May 13, 2008 at 06:34:49PM -0700, Andrew Lentvorski wrote:
However, they often fall afoul of the same problems as microkernels,
poor performance.
If you listen to Tanenbaum, he'll say that may have been true in the past.
But, now CPUs are so much faster that all is sunshine with microkernels.
If you listen to Linus, he'll say the whole idea with microkernels is still
retarded.
They *both* should know. Hard to know who is right now.
Tanenbaum, but not through effort of his.
First, what do you think a paravirtualized kernel is? It's just a
kernel running on a microkernel with a fancy marketing name.
Second, the L4 microkernel folks have shown that you can do Linux on a
microkernel substrate with about a 10% penalty (see L4Linux). That's
your existence proof. DragonFly BSD also appears to be going this way
as well and doesn't seem to be suffering too badly, either.
Third, development forces are pulling things toward microkernels just
like all processors got pulled toward RISC even if they didn't call it
that. There is simply an upper complexity limit to subsystems, and you
have to break that apart with an isolated, non-sharing API eventually.
It's easier to test; it's easier to understand; it's easier to debug.
See the difference between implementing ZFS under FreeBSD vs. Linux.
Fourth, when you start trying to add security capabilities to different
OS resources, they can't be all munged together. This is another force
pulling things toward isolated API's with much less sharing.
You have to be careful because Linus and Tanenbaum have different goals
and often talk *past* one another. Also, you have to consider that it
doesn't matter who is right. Linus and Tanenbaum will do what they will.
Personally, I think Linux is about at its complexity limit. Something
else is about to emerge. There are a bunch of pieces swirling around
that are trying to "be simpler": the Xen hypervisor, the VMWare
hypervisor, the LLVM project, L4, Coyotos, etc.
Someone is going to get hit by the bolt from the blue shortly.
-a
--
KPLUG-List@kernel-panic.org
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
