摘要
This paper deals with parallel implemeatation on distributed memory systemsof a pressure-correction projection scheme for the unsteady incompressible NavierStokes equations, the CNMT2 scheme (i.e., the Crank-Nickolson Modified Temamscheme Ⅱ), presented in [1, 2]. The key point of this work is to study parallelizationof the fast Poisson solver [2] and analyse its parallel efficiency. Various techniques,such as pipelining and canon cyclic algorithm, were used to ensure good parallel performance and scalability of the algorithm. The algorithm has been implemented using MPI message passing environment and numerical tests have been carried out on various computers, including the home made Dawn-1000 MPP system and workstation clusters.
This paper deals with parallel implemeatation on distributed memory systemsof a pressure-correction projection scheme for the unsteady incompressible NavierStokes equations, the CNMT2 scheme (i.e., the Crank-Nickolson Modified Temamscheme Ⅱ), presented in [1, 2]. The key point of this work is to study parallelizationof the fast Poisson solver [2] and analyse its parallel efficiency. Various techniques,such as pipelining and canon cyclic algorithm, were used to ensure good parallel performance and scalability of the algorithm. The algorithm has been implemented using MPI message passing environment and numerical tests have been carried out on various computers, including the home made Dawn-1000 MPP system and workstation clusters.
出处
《计算数学》
CSCD
北大核心
1998年第3期325-336,共12页
Mathematica Numerica Sinica
基金
国家自然科学基金!19772056
八六三资助
