Re: Microkernels (was Re: Sparc port?)
"M.C. Vernon" <email@example.com> writes:
> I (and some other Cambridge people) am of the opinion that the current
> hurd doesn't get the maximum benefit out of it's microkernel design, and
> one of the reasons is that its microkernel is not very micro.
No argument here!
However, the point of a kernel-based design is not that the "kernel is
small" like some mantra, it's that there are actual user benefits
available. The mere fact that the kernel is largeish isn't so bad, as
long as there aren't other negative effects.
> Sure, Mach is smaller than linux, but it's still pretty big. I think
> this is because it tries to do too much: things like IDE, SCSI and
> so on should IMHO be servers: the microkernel should do as little as
> possible (in the way networking is handling, for example).
Disk device drivers can be moved outside of the kernel; this
experiment was already done with great success in Mach. It's not
anywhere near my top priority, but it would be a nice thing.
> I understand it would be rather a lot of work to re-implement mach
> in this manner (though I dare say most of the existing code could be
> kept/moved into hurd servers), but I think it might be worth it in
> the long run....
It may well be that what we do instead is use a different, faster,