[unios] Re: No kernel

Anders Petersson anders.petersson@mbox320.swipnet.se
Mon, 21 Dec 1998 19:05:01 +0100


From: Anders Petersson <anders.petersson@mbox320.swipnet.se>

>From: Tril <dem@tunes.org>
>On Thu, 17 Dec 1998, Pat Wendorf wrote:

>> The working model, as I understand it, at this point is totally abstracted
>> hardware that handles all the tasking and memory management.  Objects
are handled
>> by other objects loaded at boot time (such as a system abstract).  Each
object
>> interacts with each other object with small IPC code.  Access to objects is
>> controlled by a security object within the system abstract.  I guess you
could say
>> the model runs in flat mode on an intel processor (I'm not as
knowledgeable about
>> this as some of the other members).  As I understand from Pieter's
model, the
>> whole object hierarchy is implemented in tree form, where top level
objects have
>> the most access, and can grant or deny it as such.  

As Pieter said, the model is "mine" if someone's.

>Could someone write this up in a document and put it on the web page?  I
>feel uncomfortable critiquing a second-hand description of some model.  If
>Pieter would be the one to do it, and he's away, I don't mind waiting for
>him to get back.  I have plenty of other things to do.

I should write a doc about it. The problem is that I feel not everything is
sufficiently worked out yet. I'll have it done anyway... adding some things
that's not discussed, and leave out other things to be specified later.

>> I don't believe we are looking
>> at a kernel at this point, just free floating objects intercommunicating in
>> defined channels.
>[and moved up from later on]
>> Agreed.  We are stating flexibly as the #1 priority of the system
>> design, and will keep that in mind.
>
>It seems two other members of your list disagree with you.
>[Anders Petersson] I want a kernel. ;)
>[Pieter] Ok, me to. It's faster.

I wrote that, but I have to admit I don't agree with myself. :) If it
prooves that a no-kernel model would be better, fine for me. Actually,
recently I've come to some conclusions that further decrease the work done
by the kernel.

>Am I to interpret this discussion that you want a minimal kernel, with
>most everything done by objects?  I'm sorry if I'm off track here, I just
>joined the list.  I will go back and read the archives to catch up soon. 

That's correct, as much as possible "objecterized".

>> > What you said sounds more like a microkernel, because you imply the
>> > objects are running in user space.  However, each resource being
>> > abstracted by a separate object isn't exactly a microkernel either.  It's
>> > more like what we're doing in TUNES:
>> >
>> > In the TUNES project, we're not using EITHER Emmanuel Marty's "no-kernel"
>> > NOR a microkernel.  Every aspect of the system is abstracted by an
object,
>> > yes.  But objects don't run in "user mode" or "kernel mode", because
>> > objects aren't tied to a specific low-level representation.  The
>> > implementation of any object or group of objects changes dynamically,
>> > while being verified to conform to its specification.  This design is
>> > extremely flexible, and can be stable if the programmer and user both
>> > desire it for a particular object.
>> 
>> What do you mean? Do the objects jump in and out of different rung levels?
>
>They might.. but that's not the point.  What happens is that the user
>controls what objects become part of (integrated with) the system. 

Hmm... in my model all objects are integrated in the same way. How could
some objects be integrated and some not? What's the advantage?

>Objects have full specification (kind of like source code, except leaving
>out all implementation specific details.  Implementation hints can be
>provided, or even full implementations, but these can be overridden).  As
>long as the specification is followed the system (with the user's help) 
>can do whatever it wants with the underlying representations. 

The sounds somewhat resembling of our way, using 'interfaces'.

>Optimizations often cross object boundaries, according to high level
>constrants (like resources available, and user preferences).
>
>The idea is an object is just what the user sees.  Underneath, objects or
>groups of objects may be replaced by other objects in other arrangements,
>as long as they act the same.  The original objects will be retained for
>the user to interact with, although the underlying ones can be accessed
>(and interacted with) if desired.

This differs from mOS (our/my model). The user sees the same things as the
system sees. The object model is maintained by individual modules (objects
of course), that implement objects in their own way. The kernel does not
interfere.

binEng



------------------------------------------------------------------------
To unsubscribe from this mailing list, or to change your subscription
to digest, go to the ONElist web site, at http://www.onelist.com and
select the User Center link from the menu bar on the left.
------------------------------------------------------------------------
UniOS Group
http://members.xoom.com/unios