A scalable, decentralized coordination infrastructure for grid environments
Manish Parashar, Zhen Li · 2007
While Grid computing is rapidly emerging as the dominant paradigm for distributed problem solving for a wide range of application domains, the heterogeneity, dynamism, and uncertainty of Grid environments result in significant application coordination challenges. A key challenge is managing the runtime dependencies and interactions among the elements. These dependencies and interactions can be complex and various and both system entities and interactions between them can be ad hoc and opportunistic. As a result, realizing these coordinations becomes extremely challenging. This research investigates a shared-space based decentralized architecture model for addressing scalable and robust coordination for Grid applications. This model employs fully decentralized architecture and provides a global virtual shared-space abstraction that can be associatively accessed by all peers in the system. In this research, we design and develop Comet coordination infrastructure to demonstrate the conceptual architecture model. The architecture of the Comet is based on a content-based distributed hash table, which employs a locality preserving mapping to map the multi-dimensional information space used by the coordinating entities to the one-dimensional peer node index space. The resulting shared-space maintains content locality and guarantees that content-based information queries are delivered with bounded costs. The key contribution of this thesis is a conceptual architecture model and an implementation infrastructure for realizing coordination abstractions that support dynamic, scalable, and asynchronous application interactions on wide-area Grid environments. The developed architecture model and the Comet infrastructure are used to support coordination and computation in Grid environments. Two prototype systems have been implemented and evaluated. The first prototype, CometG, provides a decentralized (peer-to-peer) computational infrastructure that extends Desktop Grids to support parallel asynchronous applications. CometG constructs decentralized coordination spaces and programming abstractions for parallel asynchronous iterative computations and asynchronous formulation of the replica exchange algorithm for molecular dynamics applications. The second prototype, Rudder coordination framework provides agent abstractions and coordination protocols for supporting dynamic composition of Grid applications. Experimental evaluations of these prototypes demonstrate the flexibility, scalability and effectiveness of the infrastructure, as well as its ability to support complex coordination requirements of Grid applications.