Spatial Indexing of Large-Scale Geo-Referenced Point Data on GPGPUs Using Parallel Primitives
Department of Computer Science, The City College of the City University of New York, New York, NY, 10031
The City College of the City University of New York, 2012
@article{zhang2012spatial,
title={Spatial Indexing of Large-Scale Geo-Referenced Point Data on GPGPUs Using Parallel Primitives},
author={Zhang, J. and Gruenwald, L.},
year={2012}
}
Modern positioning and locating technologies, e.g., GPS, have generated huge amounts of geo-referenced point data that are crucial to understand environmental and social-economic phenomena. Unfortunately, traditional disk-resident databases are inefficient in handling large-scale point data. In this study, we propose to utilize the massive data parallel processing power of General Purpose computing on Graphics Processing Units (GPGPUs) technologies to index large-scale geo-referenced point data by using parallel primitives for efficiency, simplicity and portability. We have developed a CSPT-P (Constrained Space Partitioning tree for Point data) tree indexing structure that is suitable for parallel construction. Experiment results using a New York City (NYC) taxi trip dataset with nearly 170 million taxi pickup locations have demonstrated a 23X speedup on an Nvidia Quadro 6000 device over a serial CPU implementation on an Intel XEON E5405 processor.
April 12, 2012 by hgpu
