Re: Design goals

[olafBuddenhagen]

Hi,

> Here's the list of design goals I promised, sorted by priority.
[...]
> stability
> robustness

This doesn't mean anything. *Every* system will make a certain level of
stability/robustness essential (even if some do not show ;-) ) --
otherwise, the system is just unusable. The real question is *what* this
level should be.

If you give stability/robustness as an absolute goal, and declare it
essential, that would mean you want to strive for perfect
stability/robustness, making all other goals subordinate to that. I'm
pretty sure that's not what you mean.

> resource accountability
> confinement
> support for most legacy applications
> persistent sessions for users
> --------- Anything above this line is essential --------------
> confinement with endogenous verification
> soft real time
> setting diverse resource distribution policies
> persistence

As pointed out before, this is a mechanism, not a goal. What you
probably mean is the ability to recover in exactly the same state after
(possibly unexpected) reboot. Various mechanisms could achieve this, and
I'm not sure they all fit unter the "persistance" term.

> no ACLs

This doesn't seem like a real goal either. It would be a means to
achieve certain security goals easily. (In exchange for giving up a
familar concept.)

> --------- Anything below this line is optional ---------------
> small memory footprint
> support for all POSIX applications
> hard real time

Generally, your categorization seems quite appalling to me.

The original Hurd set out with exactly one goal (user extensibility)
distinguishing it from existing mainstream systems, and didn't get
finished in 15 years.

Your list places something like five features not existing in current
mainstream systems in the essential and mandatory sections. Even with
all of Shapiro's experience, I very much doubt this is realistic.

Unless of course you want to enhance the existing EROS/Coyotos, instead
of designing a new system.

Which leads me directly to another issue: Your list has absolutely
nothing "hurdish" about it. If this is really a definitive statement of
your goals, I fail to see why you would want a different system, instead
of going with EROS/Coyotos -- after all, it has all the features you
listed... (Except for the optional ones, of course.)

I generally consider your priorization quite strange. Which just
demonstrates that an open-for-everything wishlist call is really
pointless and counterproductive: Everyone will come up with other ideas,
and there is not chance for any agreement.

IMHO, it's much more useful to try to restate the *origial* Hurd goals
-- after all, that's what we have in common, what has brought us
together, even if it seems some have forgotten it; and starting from
there, very very *very* carefully consider, which new goals we want to
add on top if it -- if any.

-antrik-