摘要
本文提出了面向CPU+GPU异构环境的数据密集型矢量多边形地理大数据并行计算框架(PFGAP)。PFGAP将数据密集型矢量多边形地理大数据的并行计算分解为算子、数据、粒度、并行环境及任务调度5个模块,分别设计相应的负载均衡并行计算策略;通过封装并行计算实现细节及数据密集型多边形算子的快速并行化。试验采用多边形三角剖分、栅格化及投影变换作为测试算例,采用土地利用数据作为测试数据,在不同类型的并行环境中计算并行效率。结果表明,PFGAP能很好地适用于不同类型的数据集、算子及并行计算环境。利用PFGAP实现的并行算法显著地降低了串行执行时间,取得了40.03的最优并行加速比。试验还分别测试了各个模块涉及的并行策略,结果表明取得的并行效率优于现有并行策略。
In this study,we present a parallel framework for data⁃intensive geospatial analysis on large⁃scale vector polygons over hybrid CPUs and GPUs(PFGAP).We consider workload balance in terms of operator,data,granularity,parallel environment,and task scheduling,respectively.These modules constitute the PFGAP and the parallel implementation details are encapsulated.Through applying the PFGAP,the parallel version of a serial algorithm can be easily achieved with a proper degree of workload balance.The typical polygon triangulation,polygon rasterization,and projection transformation algorithms are employed as testing algorithms,and land⁃use datasets are used as testing datasets.Results show that the implemented parallel algorithms reduce significantly the serial execution time,achieving optimal speedup ratio of 40.03.In addition,the parallel strategies involved in each module are evaluated,showing better effectiveness with conventional ones.
作者
徐云耘
周琛
李满春
XU Yunyun;ZHOU Chen;LI Manchun(School of Geography and Ocean Science,Nanjing University,Nanjing 210023,China)
出处
《测绘通报》
CSCD
北大核心
2022年第5期110-119,共10页
Bulletin of Surveying and Mapping
基金
国家重点研发计划(2017YFB0504205)
国家自然科学基金(41901318)。
作者简介
徐云耘(1997-),女,硕士生,研究方向为地理大数据时空挖掘与分析。E⁃mail:yyxu.nju@gmail.com;通信作者:周琛。E⁃mail:chenzhou@nju.edu.cn。
