Re: kernel for Hurd

[olafBuddenhagen]

Hi,

On Tue, Sep 23, 2008 at 07:19:31AM +0200, Arne Babenhauserheide wrote:
> Am Dienstag 23 September 2008 02:06:18 schrieb
> olafBuddenhagen@gmx.net:

> > AFAIK there is nothing else presently going on regarding new kernels
> > for the Hurd. All current development still happens on the
> > Mach-based version.
> 
> Also nothing with the L4 kernel? 
> 
> The L4 sounds especially interesting to me, because my William
> Stallings "Operating Systems" says that it's been proven (or shown -
> I'm not perfectly sure about the word here) that extremely small
> kernels like L4 can archieve the same speed as monolithic kernels. 

It has been proven that systems based on small kernels like L4 can be
nearly as fast as monolithic systems in certain scenarios --
microbenchmarks, research systems, embedded systems. This doesn't mean
that it scales to a production-ready, general purpose operating
system... Nobody can really tell, as to the best of my knowledge, such
system simply does not exist.

Anyways, there is general agreement that small kernels like L4 are the
way to go when building an efficient multiserver microkernel system. L4
however did go too far when ripping stuff out: It turns out that
protected IPC in the kernel is a *must* for a general-purpose system;
there is no method know to move this to user space in a sane fasion. (As
Shapiro put it, it has been shown to be possible with L4, but only at
prohibitive cost...) The Hurd/L4 architects learned that the hard way.

Some new L4 variants (L4ng, L4.sec, seL4) do add protected IPC in
various forms, and thus *could* be candidates for a Hurd port. In fact,
when the Hurd/L4 architects realized the original L4 is unsuitable, they
did look at L4ng and L4.sec for a while. However, the L4 folks were
totally new to that, still in the experimenting phase. Thus Marcus and
Neal tried to go with Coyotos instead, hoping to get a mature solution.
But as I already said, it turned out that this is also not quite the
right thing for the Hurd -- and so finally both Marcus and Neal started
writing their own kernels. Marcus lost interest in the meantime, leaving
Neal's Viengoos as the only active effort in that direction.

-antrik-