On 05/05/07, Evaldo Gardenali <evaldo@gardenali.biz> wrote:
> Hi
>
> 1. Yes, feasible, but for what?

> 2. Not sure
>
> "If both things are possible, a lot of possibilities can be done on it!"
>
> Like what?

Several things:

1. Easier debugging (I know we can use ddb instead of userland gdb,
but there are a lot of great front-end debuggers that work with gdb
and make the debugging a lot easier.

2. Faster kernel development/testing; we can modify a line, compile
and test the kernel on the fly (=E0 la DragonFly VKernel).

3. Isolation, the same idea as using Xen, Qemu or VMware; but taking
into account that NetBSD would run on top of NetBSD, I think a lot of
performance improvements and improved resource management can be done.
Let's say, a lot of syscalls can be performed directly by the main
kernel with no user_mode-kernel_mode translation; a lot of shared
libraries could be shared between NetBSD "instances", I dunno.

4. Ok, maybe vax or hpcmips architectures do not need (and wouldn't
use) this kind of virtualization, but Intel, ARM or PowerPC would do.

I am not contrary to the NetBSD dom0, domu implementation (they added
a lot of light into NetBSD); but I do not consider futile implementing
this "NetBSD on NetBSD" stuff