Zeus: An object-oriented distributed operating system for reliable applications

This paper presents the principles followed in designing Zeus, an object-oriented distributed operating system designed to study integration of recovery mechanisms into the designs of distributed command and control systems. The primary goal of the Zeus design is to define reliable object management functions for distributed command and control systems and to evaluate the performance and the correctness of the recovery mechanisms for these functions. Therefore, no implementation of this design currently exists. The user provided functions support definition of object types, creation of objects, and updating of distributed objects using atomic transactions. We are currently evaluating the performance characteristics of this design using simulation models and proving the correctness of the recovery mechanisms using formal methods based on Gypsy language [AKER83], events and state transition based models [TRIP83b], and simulation models. To achieve these goals we have refined the Zeus design to a significantly detailed level. To date we have explored this design only from the viewpoint of these goals. Several research problems necessary to implement this system remain unexplored. For example, a linguistic mechanism is needed to introduce object type definitions into the system and to define processes and transactions. A distributed operating system for highly reliable applications must provide 1) recovery mechanisms that are transparent to the application developers and 2) naming mechanisms that make the physical distribution of objects and functions transparent to the application programmer. The second feature is important to make development of distributed software no more difficult than the development of conventional software systems. The Zeus design has made a significant contribution In this direction. Other systems have integrated these two concepts in their designs, however they typically limit object management to the file storage level. To date, Argus [LISK82] is the only other system which provides a set of general mechanisms for reliable management of distributed objects of any type. Zeus provides these mechanisms and addresses several other issues such as object relocation, authentication and object protection, not included in the Argus design. Another novel feature in Zeus is the integration of the conventional database management functions into the operating system object management functions. This is important because most of the todays popular operating systems do not provide efficient machanisms for database applications [STON81]. Even with respect to its recovery model, the Zeus design differs significantly from other known designs.