GPGPU-based Processing And Analysis Of Large-scale Geographic Data

Abstract:With the rapid development of geographic information systems technology, remote sensing technology, computer technology and information technology, there are more use of the combination of technologies. The means of this combination can extract useful information from various real time data source. It can provide decision support for the resource management configuration, urban planning and management, land information management, ecological environment management, infrastructure construction, transportation planning, and so on. However, for the processing of huge amounts of data, there will be several hundred to several thousand times increase of the calculation amount, with respect to the normal processing. It’s not only a serious test of the data processing capabilities of the computer, but also a test of effectiveness of algorithm design in the processing of massive data. High performance computing technology is advancing by leaps and bounds, which brings a new direction to vast amounts of data processing work. Multi-core CPU technology and graphics processor (GPU) technology with the increasingly programmable and efficient computing power promote huge changes in the approach, which changed from previously CPU side programming, to CPU and GPU heterogeneous programming gradually, and to further development of the distributed processing and cloud computing. The changes in the way of the processing lead to the improvement of the processing efficiency, and calculation has changed from single computer to multiple computers. Simultaneously it achieves a balanced allocation of the transaction, and effectively uses a variety of computer resources in order to improve the processing efficiency. The essay suggests a study in connection with GPU parallel computing technology, and explores how to take advantage of the GPU memory characteristics to finish massive data processing tasks, in order to achieve better acceleration effect.There are several aspects of the work done in this article:1. To the segmentation and scheduling problems of the massive remote sensing image data, we put forward a rapid processing solution of image data based on CPU and GPU heterogeneous programming, and explore different GPU memory processing of remote sensing data resampling. When doing GPU parallel based image processing, we should take into consideration of the algorithm design and division of processing tasks, reasonably divide threads, and achieve parallel execution optimization, memory optimization and instructions optimization to improve overall processing efficiency. We describe Compute Unified Device Architecture (CUDA) general purpose computing model framework and its characteristics, and accelerate remote sensing image data resampling using CUDA.2. We propose a GPU based processing numerical problems in the spatial data clustering. Taking bipartite graph clustering algorithm for example, we explore appropriate parallel processing approach based on the characteristics of clustering in numerical calculation, as well as GPU parallel architecture and hardware features. We apply global memory and shared memory to accelerate technology to improve the efficiency of data processing. The experimental results show a significant improvement in the efficiency of GPU parallel computing.