StarOs, Medusa (very applicable name )

Syed Adnan (sadnan@ucsd.edu)
Thu, 20 Apr 2000 01:33:01 -0700

Name: Syed Adnan
Date : 4-19-00

StarOS, a Multiprocessor Operating System
for the support of Task Forces.

Advances in technologies making prices cheaper for processors, and idea behind distributed system is to construct a powerful, cost -effective computer using less expensive processors. We could benefits from increased-performance, better reliability and modular component based system that can be expanded easily. StarOS is an attempt to achieve all these goals using a distributed hardware (50 processors). It is a message-based, object-oriented operating system design to support large collection of concurrently executing processes cooperating to accomplish one task. These collections of processes are referred as task forces in this paper.
Approach for this operating system to distribute the work among processes solving very small part over all work in a task force. Hardware provides the primary communication among processors that all processors can access primary memory of the system. Since StarOS is an object-oriented system much like Hydra, all information is encoded in objects, and processors make use of functions and capabilities in objects to access information and accomplish tasks. StarOS, also make use of messages to allow concurrency and parallelism in the system. With parallelism, system can achieve its goal of increasing performance. Reliability is not difficult to achieve that failed processes can be replace by other processes, and it is easy to add more hardware to expand the system since all processors have primary memory available to them.
In theory this methodology seems sound and attractive, but distribution of a problem in itself is a difficult task. Paper is very well written, it is easy to understand, but I wonder that why they didn't talk much about security that much?
In my opinion, idea of using many different processors to solve a problem is a very great one, but hardware system needs to be changed. Networked, clusters of computer the wave of the future. Though we can use many ideas from this paper, I would use cluster of computers to solve a distributed problem rather having clusters of proessors in a system.

Name: Syed Adnan
Date: 4-19-00
Medusa: An Experiment in Distributed Operating System Structure

Medusa is an operating for CM architecture discussed in StarOS. Medusa's goals are to take advantage of architectural featuers to produce a modular, robust, and efficient system. Medusa is a multi-user operating system, and just like StarOS, Medusa also attempts to capitalize on the CM distributed architecture. Medusa emphasizes on structure. It does not use group of processes working on very small task to accomplish a bigger task, instead it uses utilities that communicates with each other using messaging provided by the system.
In order to achieve modularity in this system, they decided to have large number of small cooperating subcomponents independent of each other. For reliability, system should be able to response to the change in its environment, and these changes could be workload to hardware failures. Their approach, for performance, is to provide structure to operating system that reflect underlying hardware, so program can run with same efficiency using Medusa as they could on the bare hardware.
Medusa's processors also have any memory location in the system, but it could cost depending on the location of the memory. However, Medusa has discarded the notion of having operating system code executed in any point of the system. Processors are restricted that processor can only execute a utility, if it can do so locally. Messages are being used for function activation, however. No processor is self sufficient, and no processor is critical to the operation of the system. Medusa introduces exception handling which is very interesting as well.
Medusa is quiet different from StarOS, but does have trade offs to produce highly distributed operating system structure that closely matched the hardware. Task force in medusa is more restrictive than in StarOS, but does support for co scheduling and exception handling.
Though there are differences between two systems, their goal of using hardware to solve problem efficiently. I still think that clustering of computers and writing programs to distribute work would be step ahead same direction as Medusa and StarOS.