Re: A simple question

[olafBuddenhagen]

Hi,

On Thu, Apr 13, 2006 at 10:27:40AM +0200, Bas Wijnen wrote:
> On Thu, Apr 13, 2006 at 02:39:08AM +0200, address@hidden
> wrote:

> > The desire to create a completely different system only came with
> > Shapiro's influence, way after microkernel reevaluation was already
> > under way.
> 
> That's not how I remember it.

I guess you can't remember it, because most likely you missed that part
completely -- nearly nothing of it showed on the mailing list.

Around last year's FOSDEM, Marcus was already expressing *considerable*
doubts whether the original Hurd/L4 design on Pistachio is actually
feasible. Shortly after that he learned about L4ng and L4.sec, and the
following months Neal and Marcus were discussing a possible slightly
changed design on top of L4ng or L4.sec with the L4 folks. (AFAIK some
exchange with Shapiro on these issues also already started.)

At the meeting with Shapiro at RMLL, Marcus and Neal hoped to resolve
some of the open issues -- but instead, Shapiro pointed out that the
existing Hurd design probably can't meet Marcus' goals (and Neal's,
though to a lesser degree I believe), and gave some suggestions how to
address various issues. (I.e. explained how it's done in his designs,
and why he believes them superior...) He also suggested the Coyotos
kernel as a possible alternative to L4.sec.

The following months, Marcus and Neal considered *both* the possiblity
of using the Coyotos kernel, and of using the EROS/Coyotos design ideas.
If you look at the mailing list discussions that started in October,
you'll see that at this point they were actually already quite firmly
set on trying both -- most of the stuff was new only to the rest of us.
(Not terribly new to me, as I already heard the basic ideas at RMLL; but
new to you and most others.)

> Well, if you change microkernel, you need a redesign.  After speaking
> with Shapiro, it seemed clear (at least to me) that pistachio couldn't
> do all the nice things that we now knew about and wanted.  So we would
> use some other microkernel, be it Coyotos or an other L4 variant.  In
> both cases we would need a redesign.

Only the low level implementation is directly dependant on the kernel;
most of the higher-level design isn't. The Coyotos kernel and L4.sec are
actually quite similar in most regards. AIUI, you could build pretty
similar systems on any of them -- including both the original Hurd
design as exists in Hurd/Mach and was planned for Hurd/L4, as well as
the upcoming quite different ngHurd design.

> > You do not seem to have much understanding of what the (existing)
> > Hurd has to offer, if you believe a kernel like Linux could do the
> > same or even come close *without* a complete rewrite.
> 
> Seen from a user perspective, it already comes close.  The cool
> features of the Hurd are not so spectacular that they'll be impossible
> on Linux.

Please don't be ridiculous. With Linux it's possible not only to
implement new filesystems in user space, but also to change/extend the
file system interfaces themself? To replace the whole networking stack?
To change the way file access and other permissions work? To set up
proxy environments for processes? To change the signal facility, the way
programs are executed, the way sockets work, the way priviliges are
propagated etc.? In general, to change/extend nearly every aspect of the
system functionality, without any admin action, without affecting other
users, and without restarting the system?

Just because most people lack imagination to think beyond FUSE (many
actually even to think that far...), doesn't mean Hurd has nothing more
to offer.

> Right, you need root access to do things with Linux, and you don't
> need it on Hurd. We consider that a very good thing.  But do you
> really think the user cares about it?  Only when things go wrong, and
> his whole computer is taken over because he did things as root, then
> he cares.  But if that was a real concern, how many people would still
> use Windows?

So you are arguing at the same time that Hurd's security features are
useless as nobody really cares about security, and that ngHurd is more
worthwhile because it improves on some security aspects?...

> It does already compete, and it's winning big time because most people
> think performance is more important than "a cool environment for
> developers".

...is winning against what, an unfinished system that hardly anybody has
an idea what it has to offer? Big deal.

Incidentally, "a cool environment for developers" is much more important
than what "most people think". Because "most people" do not improve a
system. It's the developers who do. To the average end user, it doesn't
matter at all what mechanims the core system provides. What matters is
what they can do with the system in the end. And that depends on how
well the system helps *developers* to implement functionality.

GNU/Linux didn't become popular because it was a nice system for average
end users. It became popular because it was a nice system for
developers. And only subsequently these developers started slowly making
it useful for end users.

> However, the discussions about the features should be found when
> searching for these terms, I think.  That's what I said, too. ;-)  But
> actually, I expect most of the readers to have followed it at the
> time, so they all know what it's about anyway.

The one who started this thread obviously wasn't.

It's not terribly helpful to tell people to spend a week reading up list
archives to learn about some features. You could just as well tell them
to fuck off because you can't be bothered to answer their questions...

-antrik-