Re: mach is dead...isn't it?
On Mon, Aug 05, 2002 at 06:35:56PM -0400, B. Douglas Hilton wrote:
> No way! Mach still hasn't even hit its stride. It is designed
> for massive multiprocessing scalability across thousands and
> millions of nodes. Unfortunately Hurd hasn't advanced enough
> yet for it to really go online.
I don't want to spoil the party, but the last couple of dozen
of crashes were all Mach kernel panics, and not Hurd bugs
(unless they were Hurd bugs which triggered a kernel panic).
And that was under constant load, and not due to any leaks in the Hurd.
The VM has some defects, and the IPC is dog slow, so that it doesn't
seem that user space drivers are an option in Mach (there has been some
research about that). So it seems indeed that Mach is too monolithic and
inflexible to be an ideal solution even on a single node.
> AFAIK, L4 just doesn't have this capability. It is faster in a
> uniprocessing scenario, but cannot scale in its current state.
> The L4/Hurd project is just trying to leverage existing code
> from Hurd so that they can have a native OS for their kernel.
I think you are mistaken about that. The L4 project is there to make a
microkernel design feasible at all. I am not sure what exactly you
are thinking about, maybe you are talking about NORMA. But I don't think
there is a reason why you can not implement something like that in L4
in user space, too. L4 doesn't intend to have such capabilities. But
on the other hand, it doesn't burden you with the overhead of such
capabilities when you don't need them. This optimizes the common case,
and I don't think it is worse for the rare cases than Mach either.
> Also, Hurd need not be tied to any particular one microkernel,
That's true.
Thanks,
Marcus
Reply to: