元数据管理系统的设计与实现

大规模并行数值模拟程序通常会被运行十几次甚至上百次,每次运行都会输出几十、上百、甚至上千时间步的数据,每个时间步会产生几十、上百、甚至上千个数据文件.传统的数据管理方法是使用记事本记录数值模拟结果相关的元数据信息,这种方法的缺点是大量的信息需要手工录入,同时也不方便数据查询和共享.元数据管理系统的设计目标是通过自动提取元数据来减少需要录入的元数据信息,通过集成管理与结果数据相关的各种数据来方便数据查询与共享.该系统包含5个组成部分:层次化自描述通用数据存储模型、科学计算元数据、元数据自动提取、对象数据库系统、Web界面元数据管理系统.通过用户试用,验证了元数据管理系统实现了预期的目标.

Design and Implementation of Metadata Management System

Large-scale parallel simulation often runs dozens or even hundreds of times,and it outputs dozens or hundreds or even thousands of timesteps each time,and each timestep creates dozens or hundreds or even thousands of data files.Traditional data management method uses hardcopy notebooks to record relational metadata information of simulation results.Its shortcomings are that metadata must be manually inputted,and the searching and sharing of data are difficult.For replacing manual input with automatic metadata extracting,a metadata management system is designed to make data searching and sharing easy by integrated management.It includes five components:hierarchical self-described common data storage model,scientific computing metadata,metadata automated extracting,object database system,Web interface metadata management system.Tryout of some users validates that the system realizes its goal.