[cap-talk] Need Challenge Problems

[Jed Donnelley]

At 10:02 AM 7/20/2006, Karp, Alan H wrote:
>Jed wrote:
>
>         a lot of stuff I agree with
>
>My comment was only that I believe you can start within a machine and
>extend the model outward to the network,

You certainly can - e.g. DCCS, Mach, others?

>an approach you claimed was doomed to failure.

And I still think it is difficult to the extent of being impossible.
Basically I believe that because there are enough, what shall
I call them? - "quirks" that one can put into a single machine
(shared memory) system that seem like good ideas at the time,
but that when you get to the network extension they end up
showing themselves for what they are - opportunities that had
no good justification but then become problems when you get
to the network level.  The way RATS mapped process memory
with an invocation on a file capability is an example.  I expect
I could find examples in about any classical capability system.

I argue that if you extend such a system to a network for
of communicable capability then you get what that looks like,
a specific single machine shared memory system extended
to a network.

On the other hand if you develop a communicable permission
token for a network (e.g. along the lines of Tyler's YURLs),
you have a mechanism that can be effectively standardized
at the network level and (I argue) can work just as effectively
at the OS and language (machine) level.  No "quirk"s.

>I believe we ended up with a simpler system because
>of the approach we took.

In that case I would suggest using the network extension mechanism
you developed instead of YURLs (or whatever standard hopefully
emerges) on the Internet.  Still seem like the simpler solution
when considered in that context?

>Client Utility didn't fail for technical reasons.  HP simply decided to
>exit the middleware business.

Ditto every other capability system developed at the OS/machine
(shared memory) level.  I expect it will never be otherwise.  I still
have hope for an accepted network paradigm that can be extended
to the OS and language level - though I admit even that is thin.

Hope springs eternal.  I guess we just have hope in different areas
based on my past experience.  My hope lies in the emergence of
network standards (Go Tyler!).  It sounds like yours is in another
instance of something like Client Utility that will have enough success
to extend to the network.

I wish us both luck!!!

--Jed http://www.webstart.com/jed/