The Research Of Parallel Methods On Raster Analysis Under Multi-core Or Many-core Environments

In recent years,the rapid development of geographic information technology has been widely used,while acquisition of spatial information technology makes a spurt of progress,which make geospatial data grow with each passing day,especially in the raster spatial data,leading to that GIS raster data analysis is facing enormous pressure of calculation.The emergence of new hardware architecture of single nucleus makes that the parallel computing technology is no longer confined within the scope of supercomputers and the parallel computing technology is more and more popular.Application of parallel computing technology to achieve parallel processing of raster data analysis under the conditions of massive data,can effectively improve the computational efficiency.This paper is a parallelization method research of raster data analysis in GIS spatial analysis.Taking the terrain factor analysis and regional terrain contour modeling in terrain analysis these two kinds of raster analysis as the examples,combining data parallelism,task parallelism,double parallel and other different parallel partitioning rules,this paper explores the design and implementation of all kinds of parallel plans under multi-core CPU and many-core GPU computing environment.Finally,the paper analyses the applicability of all kinds of parallel plan,through the experimental comparison of application efficiency and advantages and disadvantages of different methods for different parallel raster data analysis methods,and eventually provide reference and research directions.The main contents of the paper are shown as follows:(1)Based on the overall object of the parallelism design of the raster data analysis,the paper has a research on the parallel feasibility of the typical parallel and the applicability of the parallel model in raster data analysis.(2)Based on multi-core shared memory model and windows multi-threading technology,the paper has a research on the parallel design and implement on several typical raster data analysis methods under different parallel models.Moreover,the paper has an optimized design on the previous parallel plan using the production-consumption model and thread pool technology.(3)Based on many-core stream processor model and CUDA programming model,the paper has a research on the parallel design and implement on one typical calculation problem that there are a lot of high independence calculations in the raster data analysis.According to the difference in the use of memory,this paper puts forward a number of parallel designs.(4)Based on the efficiency differences between different parallel calculation methods applied in the same raster data analysis,the paper attempts to construct CPU/GPU heterogeneous parallel computing model,thus providing reference for creating universal CPU/GPU heterogeneous parallel computing model of raster data analysis.In summary,this study is an attempt to apply a variety of parallel computing methods in multi-core and many-core environments to several typical raster data analysis.The results show that the parallel computing technology has good acceleration effect,but there is a gap between the acceleration of different parallel methods applying to different raster data analysis.Some raster data analysis which has a lot of logic complexity steps is more suitable for multi-core shared memory model,while some raster analysis that has a lot of independent raster operators is more suitable for many-core stream processor model.In addition,in some raster data analysis,heterogeneous parallel computing model has higher efficacy than traditional single parallel pattern.