MOLECULAR ACTIVITIES: A STRUCTURING MECHANISM FOR DISTRIBUTED COMPUTATION (ATOMIC ACTIONS, SYSTEMS, FAULT TOLERANCE

The maintenance of the integrity of shared data in a distributed system represents a major issue that has to be addressed in order to realize the potential of distributed systems for computational efficiency and reliability. The traditional approach to this problem is based on the use of atomic actions for structuring distributed computation. Although the utility of atomic actions as a structuring mechanism for achieving reliability in distributed computation is well recognized, its utility in exploiting the potential for concurrent computation in a distributed system is limited.;In this dissertation a new structuring mechanism for distributed computation called molecular activities is developed. Molecular activities provide a richer control abstraction than atomic actions and thereby provide a means for computations using them to shift their emphasis between the reliability and concurrency aspects of distributed systems as dictated by the needs of the application.;A number of approaches for implementing molecular activities in a distributed system are proposed in this thesis. We show how efficient distributed problem solving systems can be built using molecular activities thereby demonstrating the potential of molecular activities for exploiting the inherent concurrency within an application in a distributed system. The role of molecular activities in achieving fault-tolerant distributed systems is then outlined. Whereas existing schemes for achieving fault-tolerance in distributed systems treat hardware and software fault-tolerance as separate issues we propose a scheme based on a construct called a parallel conversation that uses molecular activities as an infrastructure for arriving at a uniform framework for hardware and software fault-tolerance in distributed systems.