首页 | 本学科首页   官方微博 | 高级检索  
     

面向数值模拟数据的HDF5性能优化
引用本文:沈卫超,曹立强,夏芳,宋磊.面向数值模拟数据的HDF5性能优化[J].计算机研究与发展,2012(Z1):314-318.
作者姓名:沈卫超  曹立强  夏芳  宋磊
作者单位:北京应用物理与计算数学研究所高性能计算中心
基金项目:中国工程物理研究院科学技术发展基金项目(2010B0403057)
摘    要:大规模数值模拟数据对可视化分析提出了挑战,I/O是影响可视化交互性能的重要因素.HDF5是科学计算领域广泛采用的存储格式,介绍了HDF5的抽象数据模型、数据读写流程,并使用典型数值模拟数据测试了HDF5的读性能.测试发现HDF5的数据集定位开销较大.根据数值模拟数据的数据块以整数有规律编号的特点,通过在HDF5中增加数据块视图对象来提高读性能.测试表明,该方法可显著加速数据的读取性能.

关 键 词:数值模拟  大规模数据场  HDF5  科学数据管理

HDF5 Performance Optimization for Numerical Simulation Data
Shen Weichao, Cao Liqiang, Xia Fang, and Song Lei.HDF5 Performance Optimization for Numerical Simulation Data[J].Journal of Computer Research and Development,2012(Z1):314-318.
Authors:Shen Weichao  Cao Liqiang  Xia Fang  and Song Lei
Affiliation:(High Performance Computing Center, Institute of Applied Physics and Computational Mathematics, Beijing 100094)
Abstract:Large scale simulation data challenges the ability of visualization and I/O is an important factor for visualization interactive performance. HDF5 is a common storage format in scientific computing area. The Abstract data model and the data read/write flow of HDF5 are presented. The read performance is tested by typical simulation data, and the big overhead of locating dataset is found in this test. A new object, Data Block View, is added into HDF5 to improve its read performance. Data block of simulation data is identified by regular integer, and the data block view accelerates locating speed by it. The test result indicates the method can improve read performance remarkably.
Keywords:numerical simulation  large scale dataset  HDF5  scientific data management
本文献已被 CNKI 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号