A~traet:In order to meet the demand for fast computing and retrieval of massive high-dimensional data,a high-level distributed framework for searches and computations on tree indexing structures based on map-reduce in the Hadoop environment are introduced,MRC-Tree for short.And then,two distributed KD-Tree index structures are proposed based on MRC-Tree framework,multiple KD-Trees by splitting data equally based on Map-Reduce,MKDTM for short,and one distributed KD-Tree based on Map-Reduce,OKDTM for short.Data segmentation,KD-Trees construction by parallel way,and concurrent retrieval can effectively improve the retrieval performance.Finally,the theoretical analysis and experimental results indicate that these methods are highly effective and extensible to the similarity search in high-dimensional data environment,and the OKDTM index structure performs better than the MKDTM method.%为满足海量高维数据快速计算和检索的需求,基于一个高层次的分布式树形索引结构抽象框架MRC-Tree,以及不同的KD-Tree建树方式,提出两种基于Map-Reduce机制的分布式KD-Tree索引结构构建方法,分别为MKDTM方法和OKDTM方法.通过并行对数据进行切分和建树,多个节点并发检索,可以有效地提高检索性能.理论分析和实验结果表明,基于MRC-Tree框架的分布式KD-Tree索引结构具有良好的可扩展性和较高的检索效率,且OKDTM索引结构比MKDTM具有更优良的性能.
展开▼
