时空轨迹数据的并行采集与存储  

Parallel Gathering and Storing of Spatio-temporal Trajectory Data

作　　者：夏芳[1] 沈卫超[1] 曹立强[1] 陈虹[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,北京100094

出　　处：《高性能计算技术》2012年第4期19-23,共5页

摘　　要：为解决数值模拟并行轨迹数据的输出与访问共性问题，本文实现了端到端的存储解决方案，并提出扩展的轨迹数据模型，支持复杂并行轨迹数据的表示与组织；提供直接易用的轨迹数据读写接口，屏蔽底层复杂的数据处理过程；重叠数据并行通信与合并，提供可扩展的并行数据采集技术；提供可扩展文件数组存储技术，满足多种文件访问模式需求；直接耦合可视化系统，方便用户进行轨迹数据的浏览与可视化分析。目前，研究成果已在模拟激光聚变过程的LARED集成程序中取得了良好的实际应用效果。There are some common problems existing in outputting and accessing numerical simulation parallel trajectory data. We design and implement an end-to-end data storage method to resolve them. This paper proposes a new trajectory data model which can be used to represent and organize complex parallel trajectory data. Through providing intuitionist and useable data I/O interface （ API ） , we efficiently hide the underlying complicated data processing procedures. Overlapping the data parallel collect and merge operation are made to meet the extensible application requirements of large scale data. We also realize the extensible storage technique for file arrays to support several file accessing patterns. The users can view and visual trajectory data by using JaVis visualization tool benefited from the data coupling tools provided by us At the present time, the good application effect has been achieved in the LARED simulation.

关 键 词：时空轨迹数据 并行处理 数据存储 

分 类 号：TP315[自动化与计算机技术—计算机软件与理论]