Research And Implement On Parallel Access Technology For Geographic Raster Data

With the rapid development of observation techniques, the continual growinggeospatial data makes the geographical computation data-intensive andcompute-intensive. Now as typical representative of new hardware facilities, parallelcomputing clusters and multi-core processors are becoming the mainstream gradually.They provide high performance computing power for the complex GeoComputation.However, the growing speed of I/O does not keep pace with the speed of computingperformance, which limites the massive spatial data processing capability. Therefore,the I/O performance becomes the bottlenecks of complex GeoComputation, and it is ofgreat significance to study the I/O problems in parallel GeoComputing. Focusing on theparallel access technology for geographic raster data, this dissertation studies thefollowing aspects.Firstly, the characteristics of GeoComputation, the relationship among read-writetime, dividing time and communication time in parallel processing is analyzed, and thecost models of data I/O in GeoComputation is developed. Through comprehensiveexperiments, we analyze the characteristics of distributed-collective I/O and parallel I/Ocost models, which lay theoretic and experimental foundation for further design andimplementation of parallel access methods in geographic raster data processing.Secondly, we analyze the characteristics of existing storage models of parallel fileand its applicable conditions. Considering the characteristics of complexGeoComputation, the parallel access pattern for geographic raster data is designed, theabstract data model which supports multi-source geographic raster data is proposed.Moreover the parallel access interface for geographic raster data is implemented tosupport the development of complex GeoComputation algorithms.Thirdly, based on the parallel access mode, we propose a parallel I/O architecturefor geographic raster data, design and implement four parallel I/O algorithms, andprovide a parallel programming mode of data I/O in complex GeoComputationalgorithms. Via a lot of experiments, we demonstrate that the parallel I/O model couldimprove the overall performance of the GeoComputation algorithms.Finally, under the condition of digital terrain analysis and map visualization, basedon a parallel cluster environment, the parallel access technology of geographic rasterdata is implemented. The empirical results proved the correctness and efficiency of thealgorithms.In summary, via the research on parallel I/O technology of geographic raster data,the paper improves the geographic raster data I/O efficiency, and provides technicalsupport for improving the overall performance of parallel GeoComputationsignificantly.