全球涡分辨率并行海洋模式POP在神威蓝光上的移植和应用

发布时间：2018-05-15 17:54

本文选题：神威蓝光 + 海洋环流模式　；参考：《计算机应用与软件》2014年05期

【摘要】：基于中国自行研制的超大规模并行计算机神威蓝光平台,对海洋环流模式POP进行移植和进一步优化,并进行模式和机器的并行测试。测试结果表明:0.1度分辨率在5 000核以内达到了线性加速比,随着使用线程数目的增多,虽然加速比依然增加,但是加速效率在降低,当线程数目达到24 000的时候加速比出现降低,加速效率只有12.6%;0.05度分辨率可在12 000核以内达到了线性加速比,同样随着线程数目的增多加速在增多但加速效率在降低,当线程数目达到32 000的时候加速比出现降低,加速效率只有25.7%。经过优化后的海洋环流模式POP在神威蓝光并行计算机系统上具有良好的兼容性,并行效率较高,而且具有很强的可扩展性,可用来开展超大规模的并行计算。对测试结果的进一步分析发现,发展并行性良好的线性方程组求解方案、快速高效的计算网络协议等问题是未来发展高分辨率模式中需解决的问题。
[Abstract]:Based on the super-large scale parallel computer Shenwei Blu-ray platform developed by China, the ocean circulation model (POP) is transplanted and further optimized, and the model and machine are tested in parallel. The test results show that the 0.1 degree resolution reaches a linear speedup within 5,000 kernels. As the number of threads used increases, the speedup rate still increases, but the acceleration efficiency decreases. When the number of threads reaches 24,000, the speedup ratio decreases, and the acceleration efficiency is only 12.6 degrees. The resolution of 0.05 degrees can reach a linear speedup within 12,000 kernels. Similarly, as the number of threads increases, the acceleration rate increases, but the acceleration efficiency decreases. When the number of threads reaches 32,000, the speedup decreases, and the speedup efficiency is only 25.7. The optimized ocean circulation model (POP) has good compatibility on Shenwei Blu-ray parallel computer system, high parallelism efficiency and strong extensibility, so it can be used to carry out large-scale parallel computing. Further analysis of the test results shows that the development of linear equations with good parallelism and fast and efficient computing network protocols are the problems to be solved in the future development of high resolution models.
【作者单位】：国家海洋局第一海洋研究所;江南计算技术研究所;
【基金】：国家高技术研究发展计划项目(2101AA012403) 中央级公益性科研院所基本科研业务费专项资金项目(GY02-2010G22)
【分类号】：P731.27;TP338.6

【共引文献】