Re: papers

Carnevali $ilvio (csmede@hotmail.com)
Wed, 19 Apr 2000 19:25:20 PDT

>In absolute terms...yes, it was too long :-) For that question, just
>try to keep the sentence short and concise. You should be able to
>summarize the point of the paper in a simple sentence (if I were
>writing evals, that is how I would start them...)

OK, next time I'll try to be more concise. It's pretty funny though, until
now I've been asked thrice the same question...

And here are the Evals:

StarOS

This operating system was specifically designed to support the Cm*
Multiprocessor Computer, with a particular
emphasis on cost-performance through a good use of the available resources
(CPUs). The new idea introduced
here is found in the concept of Task Force, which groups many small
processes cooperating in parallel to the
accomplishment of a specific computation. This new feature allows a process
to be split into many small
processes executing in parallel but on different CPUs of the system.

StarOS was designed in order to best suit the existing computer architecture
of Cm*, a cluster-based system
with special processors designed to manage communication needs inside and
between clusters; memory access
between processors was thus made transparent, even though a certain
communication overhead was the price to
pay.
The data was structured in objects of different types and sizes, each one
being composed of a data section and
capability section for access control, specifying the permissions and the
subset of functions that could be
used to manipulate it.
Due to the high correlation between processes in a Task Force, a secure and
fast communication mechanism was
needed; this was provided through messages sent to a special kind of
objects, Mailboxes. A scheduler would
then make it's decisions based on priority order queues as well as on the
idle state of some processes waiting
for a reply.

One of the benefits of this organization is a high flexibility that enables
it to cope with changing demand as
well as with failed process. The idea of splitting a process in many little
processes executing in parallel
on different CPUs is the main innovation presented in this paper, which also
shows how such a system needs
to be structured in order to take full advantage of that feature.
I personally think that communication between CPUs and clusters is the main
bottleneck of the system, but
unfortunately there is no experimental data supporting that supposition. The
project seems to be fucused on the
use of the system on Cm* and no mention is made about future
implementations, even though we know that StarOS
was to be replaced soon by Medusa.

Medusa

Medusa is the first evolution of the StarOS system. It was designed for the
same Cm* architecture, but with
different goals in mind.

The main goal was taking full advantage of the modular structure of the
system, to enhance Modularity, Flexibility
and Performance. For this reason, the control part of the system was divided
in disjoint utilities distributed
among the architecture, while the parallelism of executing Tasks was
enhanced.
The distribution of utilities is what differenciates Medusa from StarOS and
other systems, as it allows basic tasks
to be processor-dependent thus reducing the load of each CPU while
distributing the main features of the OS.
The whole system would then require less memory as it needn't be installed
in each PC, while being more flexible.
Communication and synchronization needs are provided via messages, pipes and
semaphores, while the process structure
introduces the new concepts of PDL and SDL simplifying interprocess
information sharing. UDLs instead are used for
communication purposes with utilities, while XDLs are introduced for data
(address) sharing needs.

Reading this paper was a good help to understand the evolution of an
Operating System as soon as hidden problems
are discovered on the previous version, independently of the architecture
which appears to be the same in this case.
At the time the paper was published the system wasn't operational yet, but a
UNIX emulator was expected to be
designed to test the full versatility of the system.

______________________________________________________
Get Your Private, Free Email at http://www.hotmail.com