Optimisation of the two-dimensional hydraulic model LISFOOD-FP for CPU architecture
Optimisation of the two-dimensional hydraulic model LISFOOD-FP for CPU architecture
复制标题
DOI：
10.1016/j.envsoft.2018.05.011
复制
发表时间：
2018-09-01
影响因子：
4.9
通讯作者：
Bates, Paul
中科院分区：
文献类型：
--
作者：
Neal, Jeffrey;Dunne, Toby;Bates, Paul
洪水淹没模型越来越多地用于各种各样的河流和沿海管理应用。然而，运行这些模型的计算工作仍然是其应用的一个重大限制。在这项研究中，LISFLOOD-FP二维洪水淹没模型的四个发展已经被记录：1）完善模型的并行化; 2）减少干细胞的计算负担; 3）减少CPU和RAM之间的数据移动;和4）矢量化的核心数值求解器。在12个测试用例中测试了这些开发在计算时间和并行效率方面的价值。对于实际的测试用例，单核性能提高了4.2倍至8.4倍，当与16个核心上的并行化相结合时，计算时间比以前的单核LISFLOOD-FP模型短34- 60倍。结果进行了比较的商业模式的背景下的样本。
Flood inundation models are increasingly used for a wide variety of river and coastal management applications. Nevertheless, the computational effort to run these models remains a substantial constraint on their application. In this study four developments to the LISFLOOD-FP 2D flood inundation model have been documented that: 1) refine the parallelisation of the model; 2) reduce the computational burden of dry cells; 3) reduce the data movements between CPU and RAM; and 4) vectorise the core numerical solver. The value of each of these developments in terms of compute time and parallel efficiency was tested on 12 test cases. For realistic test cases, improvements in single core performance of between 4.2x and 8.4x were achieved, which when combined with parallelisation on 16 cores resulted in computation times 34-60x shorter than previous LISFLOOD-FP models on one core. Results were compared to a sample of commercial models for context.