姜海燕, 尹 言, 彭川阳, 汤 亮, 曹卫星. 作物生长模型分布式并行调度方案的比较[J]. 农业工程学报, 2011, 27(6): 237-243.
    引用本文: 姜海燕, 尹 言, 彭川阳, 汤 亮, 曹卫星. 作物生长模型分布式并行调度方案的比较[J]. 农业工程学报, 2011, 27(6): 237-243.
    Jiang Haiyan, Yin Yan, Peng Chuanyang, Tang Liang, Cao Weixing. Comparison of distributed parallel scheduling schemes for crop growth model[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2011, 27(6): 237-243.
    Citation: Jiang Haiyan, Yin Yan, Peng Chuanyang, Tang Liang, Cao Weixing. Comparison of distributed parallel scheduling schemes for crop growth model[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2011, 27(6): 237-243.

    作物生长模型分布式并行调度方案的比较

    Comparison of distributed parallel scheduling schemes for crop growth model

    • 摘要: 为了提高作物生长模型的计算速度,论文提出了多种分布式并行调度方案。综合分析了田块尺度下作物生长子模型以及子模型内部组分的数据依赖关系和计算流程。以流水线技术和分治策略为基础,分别在模型组分层、子模型层和驱动数据层设计了不同的分布式并行调度方案。在WCCS2003(Windows Compute Cluster Server 2003)组成的PC集群环境下,分别采用了OpenMP、MPI_OpenMP混合以及MPI编程模型实现了多种调度方案的并行模拟。并行加速比的实验分析表明,优化后的子模型层并行调度方案,在6个双核CPUs组成的PC集群上的平均加速比可达到8.2,更接近模型并行计算加速比的预测值。在分布式集群环境下,采用基于MPI的子模型层中等粒度的并行调度方案具有更快的计算速度,更适合于作物生长模拟系统。

       

      Abstract: In order to improve the computing speed of crop growth models, multi-distributed parallel scheduling schemes were proposed. The data dependency relationships and calculation process for sub-model and sub-model’s internal components in field scale were analyzed. Based on the pipeline technology and the separate handling strategy, different distributed parallel scheduling schemes for sub-model components layer, sub-models layer and driver data layer were designed respectively. The parallel simulation scheduling schemes were realized by using programming models of OpenMP, MPI and OpenMP mixed, or MPI in the Windows Compute Cluster Server 2003 (WCCS2003) cluster environment. The results of parallel speedup experiment indicated that the optimized parallel scheme of sub-models layer could achieve average speedup to 8.2 in a PC cluster with six dual-core CPUs, which was close to the predicted value of parallel computing speedup for crop growth model. The medium granularity parallel scheduling schemes in sub-models layer based on MPI has a faster computing speed, and it is more suitable for crop growth simulation system in distributed cluster environment.

       

    /

    返回文章
    返回