The system of ensemble prediction grid portal based on GOS

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 scheduling 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.