For the task dependence of computing intensive or data intensive characteristics and the high coupling degree of parallel processing,a method of separating the execution sequence control of correlation task and processing logic of parallel al-gorithm is adopted. By means of the task decomposition approach and adaptive multi-resource subtle matching,the DEM data is used to establish the DAG task scheduling simulation of the large watershed eco-hydrology process with hundred-thousand magni-tude grids. The method performances of resolution,data scale,progress quantity and local resource manager(LRM)are evalua-ted under the condition of different resolutions by means of the multi-comparison method. The experimental results show that the DAG scheduling method of adaptive multi-resource subtle matching for task decomposition can improve the parallel performance and efficiency greatly,and has strong robustness and perfect scalability.%针对目前计算密集或数据密集特征的任务依赖和并行处理的耦合度过高,采用将关联任务的执行顺序控制与并行算法的处理逻辑相分离.该方法通过任务分解的方式和自适应的多资源精细匹配,利用DEM数据建立起十万量级栅格的大流域生态水文过程DAG任务调度模拟.在实验部分,用多重对比的方法评估在分辨率、数据规模、进程数量以及本地资源管理器(LRM)不同条件情况下该方法的性能.实验结果表明,任务分解的自适应多资源精细匹配DAG调度方法大幅度提高了并行性能和效率,具有较好的鲁棒性和扩展性.
展开▼
