基于MapReduce实现空间查询的研究  

Research on implementing spatial queries based on MapReduce

作　　者：张书彬[1,2] 韩冀中[1] 刘志勇[1] 王凯[1,2] 

机构地区：[1]中国科学院计算技术研究所,北京100190 [2]中国科学院研究生院,北京100190

出　　处：《高技术通讯》2010年第7期719-726,共8页Chinese High Technology Letters

基　　金：863计划(2009AA1Z2226);973计划(2007CB310805);国家自然科学基金(60752001)资助项目

摘　　要：为了解决原有单机空间数据库管理系统在存储能力、计算能力和可扩展能力上的不足,在分布式文件系统的基础上设计并实现了一种基于并行计算框架MapReduce的空间矢量数据管理系统Meadow,并重点讨论了利用MapReduce并行处理空间查询的数据分割方法、副本避免方法及关键算法的设计策略,最后给出了定量的分析和论证。实验表明,相对于单机Oracle Spatial,利用MapReduce并行处理空间查询的方法具有良好的性能和近似线性的加速比。实验结果也验证了MapReduce在诸如空间数据管理系统这种基于小规模集群的计算密集型应用中同样具有良好的性能。In consideration of the problem that single-node spatial DBMSs have inabilities in storage, computing and scalability, the authors designed and implemented the Meadow, a spatial vector data management system, which was built on the distributed file system and the MapReduce framework. This paper focuses on new spatial data partitioning methods and the duplication avoidance strategy in Meadow, and introduces some MapReduce-based algorithms for processing spatial queries. A quantitative analysis and an argumentation are given finally. The experiment results show that, compared with Oracle Spatial, parallel processing of spatial queries with MapReduce has a good performance and an approximate linear speedup. The performance evaluation also verifies that MapReduce performs well for computing-intensive applications in small-scale clusters.

关 键 词：空间数据管理系统 MAPREDUCE 空间查询 分布式系统 

分 类 号：P208[天文地球—地图制图学与地理信息工程]