Indexing large-scale raster geospatial data using massively parallel GPGPU computing

Abstract

Advances in geospatial technologies have generated large amounts of raster geospatial data. Massively parallel General Purpose Graphics Processing Unit (GPGPU) computing technologies have provided personal computers with tremendous computing capabilities. In this paper, we report our work on fast indexing of large-scale raster geospatial data using GPGPU computing. We have designed a cache conscious, pointerless quadtree data structure (CCQ-Tree) that has low memory footprint, is suitable for GPU indexing and allows fast mapping between main memory and hard drives. A set of algorithms have been developed and integrated to construct CCQ-Trees on GPU devices by utilizing multiple pyramid data structures and Z-order based prefix sum. Experiments on multiple 4096*4096 blocks of a global precipitation raster data have shown that CCQ-Tree indexing using an Nvidia Quadro FX3700 GPU device reduces construction times from around 9.83 seconds to 0.42 seconds (23X speedup). The experiment results have led us to project the possibility of real time indexing of global 30 arc-seconds (approximately 1km) resolution data (43200*21600) in 1-5 seconds on a personal workstation with 1-4 Nvidia next generation Fermi GPU devices.

Large-scale spatial data processing on GPUs and GPU-accelerated clusters

“…The spatial interpolation module naturally bridges point data and raster data which makes it possible to apply existing techniques for rasters for point data. Similarly the GPU-based polygon rasterization technique in [21] bridges polygons and rasters. In observing that indexing polylines and 6 https://thrust.github.io/ We have designed indexing techniques for rasters [21,22,19], points [18,26] and Minimum Bounding Boxes [26,13] using Grid-Files [26], Quadtrees [18,21,22,19] and R-Trees [13].…”

“…Similarly the GPU-based polygon rasterization technique in [21] bridges polygons and rasters. In observing that indexing polylines and 6 https://thrust.github.io/ We have designed indexing techniques for rasters [21,22,19], points [18,26] and Minimum Bounding Boxes [26,13] using Grid-Files [26], Quadtrees [18,21,22,19] and R-Trees [13]. We have also developed a GPU-based spatial join framework to join two indexed spatial datasets based on point-in-polygon tests [18], point-to-polyline distance [26], polyline-to-polyline similarity [23] with applications to spatiotemporal aggregation of large-scale taxi-trip data [18], trippurpose analysis [25], trajectory similarity query [23] and global biodiversity studies [20].…”

Accelerating Range Query Processing on R-Tree Using Graphics Processing Units

“…CCQ-tree [34] has been proposed for fast indexing of largescale raster geospatial data. CCQ-tree is similar to the CSStree in terms of that it places all nodes in a one-dimensional array to completely continuous memory allocation but is more memory efficient.…”

Speeding up large-scale geospatial polygon rasterization on GPGPUs

“…However, we refer to the works on multidimensional similarity joins [20] and density based spatial clustering [21] for examples that are relevant to spatial data processing. Our previous work on constructing min-max quadtrees from large-scale geospatial rasters has achieved a 23X speedup compared with serial CPU implementations [8]. Our recent work on decoding quadtree encoded bitplane bitmaps of large-scale geospatial rasters has achieved nearly 6X speedup when compared with a dual quadcore machine and 37X speedup compared with a single core [22].…”

“…GPGPU technologies have been successfully applied in many areas [7] including our previous work on constructing quadtrees from large-scale raster geospatial data [8]. We aim at developing a parallelization schema and an efficient implementation for GPGPU-based software rasterization and quadtree construction for large-scale polygonal geospatial data.…”