| Advances in computer networks have led to growing interest in distributed systems, enabling resource sharing. Despite the broad diversity of distributed computing projects however, resource sharing has yet to become a mainstream paradigm for compute aggregation. A number of vital challenges remain unaddressed by existing grid and metacomputing frameworks, including: (1) how to unify application deployment in spite of wide area heterogeneity, (2) how to achieve interoperability without sacrificing communication performance, and (3) how to retain reconfigurability while providing a coherent execution platform.; This Dissertation describes the H2O framework, a lightweight and reconfigurable resource sharing platform for ad-hoc heterogeneous distributed metacomputing across decentralized administrative domains. H2O approaches deployment unification through a novel, component-oriented sharing infrastructure. Providers autonomously supply component containers by executing H2O hosting daemons on their resources. Through authorization policies, they designate deployers who can further reconfigure the distributed software backplane through dynamically deployed components. Finally, end users utilize these interacting components to execute applications. Interoperability and performance issues are addressed through multi-protocol communication substrates supporting dynamic protocol switching and negotiation. These substrates provide components with consistent interaction semantics of synchronous and asynchronous method invocations, as well as distributed event notifications, that can be carried over various invocation and transport protocol stacks, fine-tuned to particular network configurations. Reconfigurability is addressed through a microkernel architecture, confining H2O core responsibilities to facilitating secure component hosting, dynamic deployment, and communication. H2O kernels virtualize underlying resources and provide a coherent deployment fabric. Middleware services can be implemented on top of this fabric through dedicated components, enabling higher-level programming frameworks that conform to desired paradigms.; Two examples of H2O-based application frameworks are described and evaluated. The first addresses the development and deployment of distributed genetic algorithms, while the other implements a component environment compliant with the Common Component Architecture (CCA) model. Experiments conducted with these frameworks illustrate the viability of H2O to provide a reconfigurable deployment and execution platform in resource-shared environments. |
