GPU parallel acceleration of transient simulations of open channel and pipe combined flows

29th IAHR Symposium on Hydraulic Machinery and Systems

Meng, W.W.^1 ; Cheng, Y.G.^1 ; Wu, J.Y.^1^2 ; Yang, Z.Y.^1 ; Shang, S.^3 ; Yang, F.^4

State Key Laboratory of Water Resources and Hydropower Engineering Science, Wuhan University, Wuhan

430072, China^1

Ministerial Key Lab of Hydraulic Machinery Transients, Ministry of Education, Wuhan University, Wuhan

430072, China^2

Zhangfeng Water Conservancy Management Company LtD, Qinshui, 048215, China^3

Construction Management Company for Chushandian Reservoir Project of Henan Province, Zhengzhou

450003, China^4

关键词: Compute Unified Device Architecture(CUDA);    Hydraulic transients;    Method of characteristics;    Open channels;    Saint Venant equation;    Speedup ratio;    Thread level parallelism;    Water transmission system;   

Others  :  https://iopscience.iop.org/article/10.1088/1755-1315/240/5/052025/pdf
DOI  :  10.1088/1755-1315/240/5/052025

来源: IOP

【 摘 要 】

Simulating the transient processes in complex water transmission system is time-consuming, and improving computational efficiency by means of parallelization on CPU clusters or even faster GPU platform is demanded. This paper proposes an approach to accelerate the transient simulations of open channel and pipe combined flows on single GPU chip. The Saint-Venant equations for open channel flows is solved by using the method of characteristics (MOC), whose inherent parallelism can be well exploited by GPU implementations in the thread-level parallelism structure of Compute Unified Device Architecture (CUDA). The sub-processes, including open channel computation, pipe flow computation and connecting boundary treatment, are implemented by different kernels. The procedures are first verified by analyzing the parallel computation efficiency of hydraulic transient processes in an open channel. Then the transient processes of a practical engineering project, which involves both open channel flow and pressurized pipe flow, are simulated. The GPU kernels are found to be memory bandwidth bounded, and the proposed single chip GPU parallel can achieve up to hundreds of speedup ratios compared to the sequential counterpart on single CPU chip.