面向大规模并行数值模拟的一种高效数据组织方式  

Efficient data organization for large scale parallel numerical simulation

作　　者：廖丽[1] 徐平均[1] 张爱清[1] 

机构地区：[1]北京应用物理与计算数学研究所高性能计算中心,北京100088

出　　处：《计算机应用》2007年第1期213-215,共3页journal of Computer Applications

基　　金：中国工程物理研究院行业科学技术预先研究基金资助项目(20050655)

摘　　要：针对现代大规模并行数值模拟数据的存储与I/O问题,提出一种高效的数据组织方式。该方式基于XML VTK格式数据I/O通用模块,输出支持随机访问、数据压缩及可移植二进制编码的并行数据文件。与传统方式相比,按照这种方式组织数据,可以降低数据的存储规模,提高可视化处理的速度。经测试,数据存储开销可降低到原存储开销的30%,数据载入到可视化软件的速度提高90%以上。且这种组织方式适于引入out-of-core技术进一步提高可视化处理效率。A new efficient data organization method was introduced in this paper which was based on a datnset I/O module developed for resolving the storing overhead problem of scientific simulation. The module exported Visualization Toolkit （VTK） XML data files for the simulation that supports random access, and portable data compression. The storing overhead was reduced significantly as well as the visualization speed was improved remarkably. The experimental results show that the data storage cost is reduced 70 percent and time consumption of loading data into the visualization system is reduced by nearly 90 percent. Another superiority of the data organization method is that out-of-core technology can be introduced into the visualization naturally to improve the data processing efficiency.

关 键 词：大规模数据场 字节序 out—of-core 可视化工具包 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]