CORC  > 清华大学
面向LICOM2的并行I/O优化
季旭 ; 武海平 ; 邹寅隆 ; 梁亚雄 ; 薛巍 ; 尹万旺 ; 林鹏飞 ; 刘海龙 ; 俞永强 ; Ji Xu ; Wu Haiping ; Zou Yinlong ; Liang Yaxiong ; Xue Wei ; Yin Wanwang ; Lin Pengfei ; Liu Hailong ; Yu Yongqiang
2016-03-30 ; 2016-03-30
关键词海洋环流模式 并行I/O优化 ADIOS OECD EU AHELO international evaluation quality assurance TP333
其他题名Parallel I/O Optimization for High Resolution Ocean Model LICOM2
中文摘要全球变暖导致自然灾害频度和强度都有所增加,发展高精度地球系统模式,对防灾减灾、促进经济发展有重要意义。随模式分辨率提高,地球系统模式的I/O性能正逐渐成为制约模式可扩展性和计算效率的瓶颈。本文针对地球系统模式的I/O问题,以代表性的海洋环流模式LICOM2为研究对象,全面分析了地球系统模式的典型I/O特点,研究通用的高并发I/O技术。本文设计实现了LICOM2系统的ADIOS并行读写方案,都取得了显著的优化效果。LICOM2系统ADIOS优化在天河1A系统内网测试,在288-6480并行度下I/O时间占比均不到4%,在6480并行度下写带宽达8.46GB/s。本文通过在天河1A上的大量测试,对提高LICOM2系统的I/O性能做了进一步的研究,不同系统配置会对I/O性能有影响。; With the global warming, natural disasters have happened more frequently and severely. It is critical to foster the research and development of the earth system model, which helps disaster prevention and economic development. As the forecast model requires higher resolution, the I/O performance has become the bottleneck of computation scalability and efficiency for the earth system model. We use the global ocean circulation model LICOM2 as a typical case, to comprehensively analyse its I/O features and find out the high performance I/O techniques for the earth system model. We have implemented the parallel I/O modules of LICOM2 by using ADIOS, which has achieved good I/O performance improvement. The I/O time of LICOM2 based on ADIOS is no more than 4% of total time with 288-6480 cores on the Tianhe-1A system. Its write bandwidth is up to 8.46GB/s on 6480 cores. we have made large amounts of tests on the Tianhe-1A system to further research on improving the I/O performance, to find out that different system settings have influence on the I/O performance.
语种中文 ; 中文
内容类型期刊论文
源URL[http://ir.lib.tsinghua.edu.cn/ir/item.do?handle=123456789/146966]  
专题清华大学
推荐引用方式
GB/T 7714
季旭,武海平,邹寅隆,等. 面向LICOM2的并行I/O优化[J],2016, 2016.
APA 季旭.,武海平.,邹寅隆.,梁亚雄.,薛巍.,...&Yu Yongqiang.(2016).面向LICOM2的并行I/O优化..
MLA 季旭,et al."面向LICOM2的并行I/O优化".(2016).
个性服务
查看访问统计
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。


©版权所有 ©2017 CSpace - Powered by CSpace