OpenCL框架下的流域径流模拟分布式调度研究
|
摘要
基于分布式水文模型的径流模拟以像元为单元进行并行计算并逐小时迭代,计算量大,故将OpenCL引入流域径流模拟中,以提高径流模拟效率。由于普通计算机显存有限,面对流域范围大、分辨率高的情况,一般以分块计算方式解决;虽然分块计算能解决显存不足问题,但频繁的显存和内存的数据交换降低了通用计算性能。该文尝试将分布式计算与通用计算相结合,将多台计算机显存资源整合,以避免频繁的显存与内存数据交换。首先将流域数据分为单台计算机能够存储和处理的子块,然后通过网络送至服务器不同的计算机上调用OpenCL计算;每执行完一次径流模拟,服务器相关节点通过网络交换计算后的边缘数据。实验证明,分布式计算和通用计算相结合能快速完成大流域的径流汇流模拟,具有一定的应用价值。
The runoff simulation based on distributed hydrological model is computed parallelly based on pixel units and iterated hourly,the amount of calculation is very huge,so OpenCL is introduced into runoff simulation to improve the efficiency.Because the video memory of common computer is limited,in the face of large watershed area and high resolution,the general solution is the block calculation.Although the block calculation can solve the problem of insufficient video memory,the frequent data exchange of video memory and internal memory reduces the performance of general-purpose computing.This paper attempts to combine distributed computing with general-purpose computing,combining the video memory of multiple computers to avoid frequent data exchange between video memory and internal memory.First,watershed data is divided into sub blocks that can be stored and processed by a single computer,then they are sent to different computers to call OpenCL to compute by network.After each execution of runoff simulation,the related nodes of server exchange will calculate marginal data by network.The experiment proves that the combination of distributed computing and general-purpose computing can complete the runoff simulation of large watershed quickly.It has definite application value.
The runoff simulation based on distributed hydrological model is computed parallelly based on pixel units and iterated hourly,the amount of calculation is very huge,so OpenCL is introduced into runoff simulation to improve the efficiency.Because the video memory of common computer is limited,in the face of large watershed area and high resolution,the general solution is the block calculation.Although the block calculation can solve the problem of insufficient video memory,the frequent data exchange of video memory and internal memory reduces the performance of general-purpose computing.This paper attempts to combine distributed computing with general-purpose computing,combining the video memory of multiple computers to avoid frequent data exchange between video memory and internal memory.First,watershed data is divided into sub blocks that can be stored and processed by a single computer,then they are sent to different computers to call OpenCL to compute by network.After each execution of runoff simulation,the related nodes of server exchange will calculate marginal data by network.The experiment proves that the combination of distributed computing and general-purpose computing can complete the runoff simulation of large watershed quickly.It has definite application value.
引文
[1]徐宗学.水文模型[M].北京:科学出版社,2009.9-10.
[2]徐宗学,程磊.分布式水文模型研究与应用进展[J].水利学报,2010,41(9):1009-1017.
[3]郑贵元,李建贵,孙伟,等.分布式水文模型研究进展[J].安徽农学通报,2017,23(10):27-33.
[4]吉田圭介,田中龍二,前野詩朗.GPUによる分流を含む洪水流計算の高速化[A].土木学会水工学委員会.水工学論文集[C].2016.59.
[5]VACONDIO R,PALUA D,MIGNOSA P.GPU-enhanced finite volume shallow water solver for fast flood simulations[J].Environmental Modelling&Software,2014,57:60-75.
[6]DULLO T,KALYANAPU A,GHAFOOR S,et al.Computational performance of a two-dimensional flood model in single and multiple GPU frameworks[A].EGU General Assembly 2015[C].2015.14739.
[7]尹灵芝,朱军,王金宏,等.GPU-CA模型下的溃坝洪水演进实时模拟与分析[J].武汉大学学报(信息科学版),2015,40(8):1123-1129.
[8]王金宏.基于GPU-CA模型的溃坝洪水实时模拟与分析[D].成都:西南交通大学,2014.
[9]SANDERS J,KANDROT E.CUDA by Example:An Introduction to General-Purpose GPU Programming[M].US:AddisonWesley Professional,2010.
[10]靳华中,柯敏毅.一种基于集群的栅格图像分割算法的研究[J].湖北工业大学学报,2006,21(4):104-107.
[11]DU P,WEBER R,LUSZCZEK P,et al.From CUDA to OpenCL:Towards a performance-portable solution for multi-platform GPU programming[J].Parallel Computing,2012,38(8):391-407.
[12]MUNSHI A,GASTER B,MATTSON T G,et al.OpenCL Programming Guide[M].US:Addison-Wesley Professional,2012.
[13]刘永和,冯锦明,徐文鹏.分布式水文模型的GPU并行化及快速模拟技术[J].水文,2015,60(4):20-26.
[14]阴桂梅,刘耀军,郭广行.云计算在分布式地理信息系统中的应用[J].山西电子技术,2010,38(6):28-29.
[15]DO H T,LIMET S,MELIN E.Parallel computing flow accumulation in large digital elevation models[J].Procedia Computer Science,2011,4(4):2277-2286.
[2]徐宗学,程磊.分布式水文模型研究与应用进展[J].水利学报,2010,41(9):1009-1017.
[3]郑贵元,李建贵,孙伟,等.分布式水文模型研究进展[J].安徽农学通报,2017,23(10):27-33.
[4]吉田圭介,田中龍二,前野詩朗.GPUによる分流を含む洪水流計算の高速化[A].土木学会水工学委員会.水工学論文集[C].2016.59.
[5]VACONDIO R,PALUA D,MIGNOSA P.GPU-enhanced finite volume shallow water solver for fast flood simulations[J].Environmental Modelling&Software,2014,57:60-75.
[6]DULLO T,KALYANAPU A,GHAFOOR S,et al.Computational performance of a two-dimensional flood model in single and multiple GPU frameworks[A].EGU General Assembly 2015[C].2015.14739.
[7]尹灵芝,朱军,王金宏,等.GPU-CA模型下的溃坝洪水演进实时模拟与分析[J].武汉大学学报(信息科学版),2015,40(8):1123-1129.
[8]王金宏.基于GPU-CA模型的溃坝洪水实时模拟与分析[D].成都:西南交通大学,2014.
[9]SANDERS J,KANDROT E.CUDA by Example:An Introduction to General-Purpose GPU Programming[M].US:AddisonWesley Professional,2010.
[10]靳华中,柯敏毅.一种基于集群的栅格图像分割算法的研究[J].湖北工业大学学报,2006,21(4):104-107.
[11]DU P,WEBER R,LUSZCZEK P,et al.From CUDA to OpenCL:Towards a performance-portable solution for multi-platform GPU programming[J].Parallel Computing,2012,38(8):391-407.
[12]MUNSHI A,GASTER B,MATTSON T G,et al.OpenCL Programming Guide[M].US:Addison-Wesley Professional,2012.
[13]刘永和,冯锦明,徐文鹏.分布式水文模型的GPU并行化及快速模拟技术[J].水文,2015,60(4):20-26.
[14]阴桂梅,刘耀军,郭广行.云计算在分布式地理信息系统中的应用[J].山西电子技术,2010,38(6):28-29.
[15]DO H T,LIMET S,MELIN E.Parallel computing flow accumulation in large digital elevation models[J].Procedia Computer Science,2011,4(4):2277-2286.