GIS data sets continue to grow at a tremedous pace, NASA EOSDIS being a succint example. Processing such large data sets requires efficient methods. In this paper we discuss issues surrounding storage and processing such data sets in a shared nothing environment. We examine several parallel R-tree structures for indexing these large spatial data sets. We especially focus on algorithms for employing the parallel R-trees in the filter phase of the parallel spatial R-tree -based join operation. We then discuss the filter phase of the join operation as relates spatial data declustering strategies, static and dynamic load balancing strategies and system scalability. We present preliminary experimental results on the join operation performed using the Digital Chart of the World Data data set on the IBM SP2 multi-computer.
展开▼
