Re: A niche for the Hurd

[olafBuddenhagen]

Hi,

On Wed, Oct 29, 2008 at 10:35:48AM +0200, Sergiu Ivanov wrote:

> Is it possible to run only *parts* of the Hurd on separate computers,
> that is, make the Hurd a kind of a network operating system. It most
> probably sounds crazy, but I know that only a microkernel OS can do
> something like that, so the Hurd may be appropriate :-)

This is actually not far off: In fact distributed systems were a (or
even *the*?) major research focus of Mach -- Mach provides
network-transparent RPC for that purpose.

I don't know of anyone ever having tried running Hurd in a distributed
fashion; but I guess it wouldn't be too hard to make it work. However, I
consider it more of a burden than a feature.

In my understanding, the network transparency is responsible for a
considerable part of the complexity (slowness) of Mach's IPC mechanism.
It also forces the network drivers to be in the kernel.

And it's not terribly useful either: Today even the smallest machines
have enough resources that there is no motivation to avoid having a
complete instance of the operating system on each node.

Distributed systems are of course still interesting at the application
level; but the general consensus nowadays seems to be that such stuff is
better implemented at a higher level than IPC primitives -- definitely
not inside a microkernel at least...

-antrik-