Abstract: Recovery logs are an important facility for fault-tolerant distributed computing that should be provided as an operating system service, l towever, a trivial adaptation of the log management component of traditional database systems is not adequate for a logging service shared by multiple resource managers. This position paper motivates operating system provided log services, and discusses some of the problems with such a service. We are implementing a new log service for the QuickSilver distributed operating system that addresses these problems.

Abstract: Distributed computing systems have received considerable attention in the last decade. Unfortunately, current research efforts are often restricted to homogeneous environments. There seems to be little attention for real-world installations, where heterogeneity is mostly the rule, rather than the exception. The Distributed Academic Computing Network Operating System, DACNOS for short, presents a general solution for running distributed applications in heterogeneous networks. The DACNOS extends local guest operating system services to provide homogeneous networking functionality. This is achieved by a virtual global object space, to which each user has access from within his native environment. This paper describes the architecture of the DACNOS network operating system, with special emphasis on its incarnation on the UNIX platform. It presents the DACNOS solutions to the problems of communication, access protection and data representation in a network of cooperating heterogeneous systems. DACNOS implementations currently exist for VM/CMS, VAX/VMS, PC-DOS, OS/2 and UNIX System V derivatives.

Abstract: The authors are concerned with fault tolerance at the level of the control programs written by the application designer. Their goal is to provide high-level architecture-independent facilities for allowing control programs running on a distributed computing system to tolerate failures of the underlying hardware and software design faults. The authors propose a model based on modular structuring of the application tasks, which permits building dependable control structures by the use of redundancy. The authors present an implementation of such an approach to fault tolerance based on the control language Syter and intended to support distributed real-time applications. This language provides various predefined operators to express sequential, synchronized, or parallel execution of processes, as well as the preemption and priority of some evolutions over others. Moreover, it offers a set of operators dedicated to fault tolerance, some of which are described. >

Abstract: We view the design of fault-tolerant computing systems as an engineering endeavor. As such, this activity requires understanding the theoretical limitations and the scope of the feasible designs. We survey the impact that various environment characteristics and design choices have on the resultant system properties. We propose a single metric—the system reliability—as an appropriate measure for exploring tradeoffs among a potentially-large design space.

Abstract: Due to the rapid development of VLSI technology, computing systems can be made much more distributed and to match the structures of the problems for various applications. Because the characteristics of distributed computing systems are significantly different from those of centralized computing systems, different a,pproaches are needed to effectively address t,he design issues related to the characteristics of software for distributed computing systems. In this pa.per, the methods currently used to develop the software for distributed computing systems are classified into three categories: dataflow-oriented, communicationoriented, and object-oriented. Their applicabilities are conpa.red, and the future trends of software design approaches are discussed. Object-oriented software design methodoiogy is identified as a more promising approach to software design for distributed computing systems.