2-DIMENSIONAL VISCOUS-FLOW COMPUTATIONS ON THE CONNECTION MACHINE - UNSTRUCTURED MESHES, UPWIND SCHEMES AND MASSIVELY PARALLEL COMPUTATIONS

被引:23
作者
FARHAT, C [1 ]
FEZOUI, L [1 ]
LANTERI, S [1 ]
机构
[1] INRIA SOPHIA ANTIPOLIS,F-06560 VALBONNE,FRANCE
关键词
D O I
10.1016/0045-7825(93)90141-J
中图分类号
T [工业技术];
学科分类号
08 ;
摘要
Here we report on our effort in simulating two-dimensional viscous flows on the Connection Machine, using a second-order accurate monotonic upwind scheme for conservation laws (MUSCL) on fully unstructured grids. The spatial approximation combines an upwind finite volume method for the discretization of the convective fluxes with a classical Galerkin finite element method for the discretization of the diffusive fluxes. The resulting semi-discrete equations are time integrated with a second-order low-storage explicit Runge-Kutta method. A communication efficient strategy for mapping thousands of processors onto an arbitrary mesh is presented and proposed as an alternative to the fast north-east-west-south (NEWS) communication mechanism, which is restricted to structured grids. Measured performance results for the simulation of low Reynolds number chaotic flows indicate that an 8K CM-2 (8192 processors) with single precision floating point arithmetic is at least as fast as one CRAY-2 processor.
引用
收藏
页码:61 / 88
页数:28
相关论文
共 20 条
[1]  
BOKHARI SH, 1981, IEEE T COMPUT, V30, P207, DOI 10.1109/TC.1981.1675756
[2]  
CHROBAK M, IN PRESS THEORET COM
[3]  
Dahl E. D., 1990, P DISTR MEM COMP C C
[4]   TRANSIENT FINITE-ELEMENT COMPUTATIONS ON 65536 PROCESSORS - THE CONNECTION MACHINE [J].
FARHAT, C ;
SOBH, N ;
PARK, KC .
INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING, 1990, 30 (01) :27-&
[5]   A SIMPLE AND EFFICIENT AUTOMATIC FEM DOMAIN DECOMPOSER [J].
FARHAT, C .
COMPUTERS & STRUCTURES, 1988, 28 (05) :579-602
[6]   ON THE MAPPING OF MASSIVELY PARALLEL PROCESSORS ONTO FINITE-ELEMENT GRAPHS [J].
FARHAT, C .
COMPUTERS & STRUCTURES, 1989, 32 (02) :347-353
[7]   A GENERAL-APPROACH TO NONLINEAR FE COMPUTATIONS ON SHARED-MEMORY MULTIPROCESSORS [J].
FARHAT, C ;
CRIVELLI, L .
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING, 1989, 72 (02) :153-171
[8]   A CLASS OF IMPLICIT UPWIND SCHEMES FOR EULER SIMULATIONS WITH UNSTRUCTURED MESHES [J].
FEZOUI, L ;
STOUFFLET, B .
JOURNAL OF COMPUTATIONAL PHYSICS, 1989, 84 (01) :174-206
[9]  
FLOWER JW, 1986, CALTECHJPL C3P292 RE
[10]  
HAMMOND S, 1991, 29TH AER SCI M REN