气候数值模式在计算网格环境下的应用

Climatic Numerical Model Applied in Computing Grid Environments

气候数值模式研究一直受到大型计算机系统计算性能和计算资源不足的制约。目前越来越多的超级计算领域,正采用网格技术来实现多个计算节点协作的分布式计算环境,以提高计算能力。本研究是气候数值模式在网格环境下的实现方案,依据数值模式数据通讯量的多少,按强关联型和弱关联型两种主要的计算形态对多个气候模式和准数据并行的数值模式在局域网网格和广域网网格环境下实验。实验发现气候数值模式在网格环境下特点:强关联型的数值模式不适合广域网络构成的网格,而弱关联型的数值模式可以在网格计算环境中取得较好的计算效果。另外,合理地组织计算节点的结构,可以有效地提高数值模式的计算效果。这些工作对气候数值模式研究在以网格计算环境为基础的下一代超级计算中的应用有借鉴作用。

The research of climatic numerical modeling has been long restricted by the insufficient resources and capabilities of the super computers. More and more supercomputing areas are using grid technology to implement distributed computing environments cooperated with multiple sites. Models with two types of computing features, fine-grain and coarse-grain forms classified by data volumes in numerical model communication were employed for grid computing case studies. We carried out experiments in several climatic models and quasi data-parallel model over both local and wide area network. We found that not all climatic models are suitable for grid computing： finegrain models on wide area network are impractical, while the efficiency of coarse-grain model is acceptable; the better computation effect can be achieved if the computing modes are organized more reasonable. This works can help the climatic researchers to develop applications on the grid platforms, which present a framework of next generation supercomputing.