GOS支持的集合预报网格门户系统

The system of ensemble prediction grid portal based on GOS

为充分利用异地分布的网络高性能资源,基于网格技术和中国国家网格中间件GOS,实现了集合预报网格门户系统.首先给出了该系统的结构和所包含的主要功能组件;然后提出了调度组件的核心设计思想,分析了集合预报流程的具体计算特征,设计了适合于不同应用场景的3种调度方式;接着给出了调度方式间的实际运行性能比较,结果表明动态竞争调度方式在资源负载不稳定时明显优于静态时间优化的提交方式;最后给出了系统的具体实现和实际部分运行界面,实践表明本系统是切实可行性的.

To exploit efficiently the distributed high performance resources across China,we implemented a grid portal for the ensemble prediction system based on the software called GOS which belongs to Chinese national grid infrastructure.In this paper,we firstly showed architecture and constituted components of this system.Then the key idea to design for the scheduing component was presented.With analyzing real computing characteristics for the process of ensemble prediction,we designed three scheduling methods adapting for different motivating cases respectively.Afterwards,we presented among these scheduling methods several comparisons on running performance,whose results showed that it was better greatly on performance for the scheduling method with dynamic competition than the scheduling method with pre-optimal makespan.In the end,the practical implemention and some running interfaces for the presented system were demonstrated,with the proof of our feasible work in the practice.