Research On Task Scheduling Strategy In Grid Environments

As a hot research topic in recent years, grid has been developed as a method to solve calculation-intensive problems of large-scale, excessive complication and those data-intensive ones. The task scheduling is the core content that the grid studies. According to certain rules and goals, grid mainly studies how to solve the key problems of the allocation of resources and implementation on the basis of unified description and organization of the grid resources and its attribution. Grid is different from parallel computing systems, which makes research task scheduling face some new challenges under grid environment.The thesis makes a deep research and analysis on the basis of grid functional characteristics, researching results and unsolved problems of grid task scheduling. To solve the existing problems, a grid task scheduling model with auto-adaptive level is designed, and its functional modules of various components are described. According to grid resource extraction under static and dynamic parameters, resources are organized by mixed-level tree structure, which provides unified information resources supportive services for scheduling strategy. Introducing the concept of resources credit, the corresponding parameters of a group constituted weighting factors are designed to conduct formal definition of resources credit and analyze the effect of various parameters on the resources credit.According to the features of grid resources, combine makespan with resources credit, a local objective function is presented, and a novel heuristic dynamic task scheduling algorithm is also proposed. The proposed algorithm, in each of the steps, in accordance with DAG critical path changes shorten makespan, makes dynamic parameters acquisition regularly and updates resource credit. The algorithm can meet user quality of service, as well as shorten makespan.By using SimGrid, a grid simulation tool, to simulate grid circumstance, simulation algorithm experiments are made. The experimental results and the analysis show that the proposed algorithm in this thesis is more applicable to the grid environment with higher efficiency.