Parallel Spatial Index Algorithm Based On Hilbert Partition

With the development of computer technology, the GIS system and technology hasmade great strides. Spatial data will be increasingly complicated, GIS model andalgorithm will also be more and more complicated, which leads to high-frequencysystem processing I/O operations and high-density calculations, its time efficiencyrequirements will also be more and more demanding. Spatial data Organization stylefrom the file system to the database schema, and then to the current relational databasemanagement constantly towards progress. The cloud-based service model andtechnology will become the inevitable trend of development of GIS technology.Based on the study of distributed data storage and parallel programmingframeworks, this paper proposed efficient spatial dataset storage frameworks based onHadoop technology. And designed the massive spatial vector’s parallel index buildingalgorithm based on Hilbert partition. Designed on the basis of the index based onMap-Reduce parallel indexing for massive parallel processing of spatial data. Bystudying of the architecture in a distributed file system HDFS storage system, weprovided technologies for the distributed storage of spatial data. Built out the HilbertR-Tree-a Hadoop-based spatial data parallel index algorithm. Conduct experiments forthe comparison and analysis to verify the performance of the Hadoop DistributedStorage. In this paper, based on the the Hilbert partition algorithm, we built a parallelspatial data index considering both the space objects aggregation and the balance of thevarious divided data storage nodes. To avoid a single or very small number of nodes"overheating" of the situation in a environment represented by Oracle Spatial when thespatial database system partition the spatial datasets based on X, Y coordinate rangepartition algorithm.By using the MapReduce programming model to solve the bulk loading problem ofthe R-tree index building we give a new way to index the vector datasets. Theexperimental results analysis shows that a near-linear performance improvement can bereached. Parallel spatial indexing based on Hilbert partition method can effectivelyreduce the building time. However, the building’s time is not the only concern of the R-tree index problem, the quality of the index are also very important. Compared withindexes built on a single processor, the MapReduce-based method generated R-treeindexes have effective been improved on generating its MBR area ranges and overlapregions. The experiments confirmed that the MapReduce method has a lot of potentialin the processing of spatial datasets, and worth in-depth study.