Resource Selection and Topology Transformation on Computational Grids

• Main Authors: Uei-Ren Chen, 陳威仁
• Other Authors: 林偉
• Format: Others
• Language: en_US
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/86500652972335697227

博士 === 國立中興大學 === 資訊科學與工程學系所 === 99 === Grid computing represents a new paradigm in distributed computing, integrating the heterogeneous computers, networks, databases, scientific instruments, and other resources managed by multiple organizations. Grid technology can provide massive computing power to support applications requiring large-scale computation or data analysis. This dissertation addresses three major topics related to computational grids. The first topic deals with the characterization of grid resources. The author defines a grid resource model and establishes a workflow to depict the grid resource topology, using a number of probabilistic and statistical techniques. These techniques are then used to approximate the characteristics of grid resources in the real world. The second topic deals with the selection of resources on computational grids. The author proposes several fundamental policies for grid resource selection and evaluates them according to their performance in scheduling DAG-like problem models. The experiments produces two significant results: the useful number of selected grid resources is strongly bounded by the implicit parallelism of tasks in the problem models; and the scheduling length can be reduced by considering both of computational and communicational costs to solve problems on computational grids. Thirdly, the author proposes a heuristic transformation algorithm for structuring a set of grid resources in arbitrary topology to form a virtual mesh. The transformation of the virtual mesh enables efficient mapping as well as multiple parallel mesh-structured computations using a given set of grid resources. Finally, the techniques associated with the three research topics presented in this dissertation provide a solid base from which to construct a problem-solving environment on computational grids.