Accord: A programming system for autonomic self-managing applications.

摘要

The increasing complexity, heterogeneity, and dynamism of emerging pervasive Grid environments and applications result in significant development and management challenges. This is primarily because application requirements and runtime behaviors depend on the runtime state and execution context and are typically not known a priori. Recently, autonomic computing has proposed solutions to address these challenges that draw inspiration from biological system. The goal of autonomic computing is to develop applications and systems that can manage themselves based on high level guidance from humans.; In this thesis, we develop the Accord programming system for autonomic self-managing applications. Accord builds on existing programming systems and extends them to (1) enable the definition of autonomic elements that encapsulates functional and non-functional specifications, rules, and mechanisms for self-management, (2) enable the formulation of self-managing applications as dynamic compositions of autonomic elements, and (3) provide a runtime infrastructure for the correct and efficient runtime execution of rules to enforce self-managing behaviors in response to changing requirements and execution context.; Three prototypes of the Accord programming system have been implemented and customized to support different classes of applications. The first prototype enables the rule-based self-management of objects and object-based parallel/distributed applications. The second prototype extends the Common Component Architecture Ccaffeine framework to enable self-managing component-based high-performance parallel/distributed scientific applications. This prototype supports both function and performance driven intra- and inter-component adaptations, and enables dynamic composition and runtime component replacement. The third prototype supports self-managing service-based applications and enables runtime adaptation of service and service interactions, and decentralized and dynamic service composition. The operation of these prototypes is illustrated using a suite of scientific applications. Experimental evaluations of the prototypes are presented.