基于HBase的海量地理空间数据的空间索引模型构建与优化

Construction and Optimization of Spatial Index Model for Massive Geospatial Data Based on HBase

传统关系型数据库在海量地理空间数据的存储与管理上面临着高并发访问规模限制、数据库扩展能力不足等困难。非关系数据库如HBase等以其强大的扩展能力与计算能力为该问题提供了新的思路与方法。空间索引模型和分布式存储模式设计是影响基于非关系数据库的海量地理空间数据的存储与查询效率的关键因素。对当前主要基于HBase的索引模型和空间数据存储设计进行了研究,设计了基于行政区划编码与矢量要素编码结合的RowKey(行键),使空间数据在HBase存储中得到很好的聚类效果,并针对要素重叠与边界划分等问题提出了一种基于四叉树-R树的改进的空间索引模型。该模型基于四叉树结构将空间数据划分为多个子网格,为每一个子网格构建R树索引,利用Hilbert(希尔伯特)曲线对子网格进行编码,并设计了基于MapReduce的并行化索引构建算法和相应的空间查询算法。经实验测试,该存储设计和空间索引模型具有较好的查询效率。

The traditional relational database is faced with difficulties in the storage and management of massive geospatial data, such as the limitation of high concurrent access scale and the lack of database expansion ability. Non-relational databases such as HBase provides new ideas and methods for this problem. Distributed storage mode and spatial index model are key factors which affect the distributed storage and query efficiency of massive geo-spatial data based on non-relational database. This paper studies the storage mode of HBase and the current spatial index model based on non-relational database, and designs a kind of RowKey based on the combination of administrative division coding and vector element coding which leads to a good clustering effect for the geo-spatial data in HBase storage. The paper improves a spatial index model based on quadtree-R tree through proposing to solve the problems of feature overlap and boundary division. A parallel index construction algorithm based on MapReduce is designed in this paper to further improve efficiency of algorithm. In the model, the index space is divided into multiple subspace based on the quadtree structure, and the subspace is encoded by Hilbert curve. The experiments have verified the model has higher time efficiency and accuracy than the existing models in spatial data storage and query.