面向新型硬件构架的新一代GIS基础并行算法研究

On Basic Geographic Parallel Algorithms of New Generation GIS for New Hardware Architectures

随着减灾应急、流域模拟、智能交通、宏观规划、区域发展等大型地学问题的不断涌现,地理信息系统(GIS)处理的数据量和计算规模不断扩大,而主流GIS仍以串行计算为基础框架,不能充分利用和发挥当前新型硬件构架(单机多核、多机多核、集群等)计算机资源的能力,难以满足实际应用的规模与高效需求。该文在分析了基础地理算法研究现状的基础上,按计算数据的关联性将基础地理算法的计算特征分为本地计算、邻域计算、区域计算和全局计算,按计算过程的资源消耗分为数据密集型、计算密集型和I/O密集型,提出了相应的并行计算策略,包括串行算法的并行改造、并行算法的性能提升和并行算法的创新设计等。进而研发了面向新型硬件构架的新一代GIS的基础地理并行计算算法库和中间件,并已集成到国产高性能GIS平台——HiGIS中,将会促进我国GIS研究、技术、系统和应用的跨越式发展。

With the continues emergence of large amount of geodata processing, such as geoscience issues on disaster reduction and emergency management, watershed simulation, intelligent transportation, macroscopic planning, and regional development, geographical information system（GIS） needs to have the ability of dealing with massive data and high computing complexity. However, the mainstream GIS software can not satisfy the new requirements on large scale and high performance, because the traditional serial computation, which have been used as the basic frame since the origin of GIS, cannot make full use of the com- puting resources of new hardware architectures. In order to solve this bottleneck problem, it＇s needed to develop a library of bas- ic geographic parallel algorithms library for new generation GIS suitable for new hardware architectures. After analysis to the research situations of basic geographic algorithm, this paper divides the calculation features of these algorithms according to its relevance with data as local computing, neighborhood computing, regional computing and global computing. Referring to the re- source consumption in the process of calculation, these algorithms are also divided into data intensive, computationally intensive, and I/O intensive types. Moreover, the authors propose corresponding parallel computing strategies,which include the transfor- mation of serial algorithm to parallel algorithm, the optimization of existing parallel algorithms, and the innovative design of parallel algorithms. Furthermore, the authors designed and developed a basic geographic parallel algorithms library and middleware for the new generation of GIS, and integrate it to a high performance GIS HiGIS. The parallel algorithm studies and middleware development introduced in this paper will facilitate the great improvement of GIS researches, technology, system and applications in China.