Implementation of an Euler/Navier-Stokes finite element algorithm onthe Connection Machine

Shapiro, Richard A.

doi:10.2514/6.1991-438

Cited by 2 publications

(1 citation statement)

References 2 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The reader is referred to articles by Mathur 1989, 1990], Belytschko et al 1990], Farhat et al 1990 and 1992], Shapiro 1991], and Johan et al 1992] for several di erent data parallel nite element algorithms. This article identi es a common set of communication primitives in the data parallel programming model which express most of the communication needs for unstructured nite element simulations.…”

Section: Introductionmentioning

confidence: 99%

Communication primitives for unstructured finite element simulations on data parallel architectures

Mathur¹,

Johnsson²

1992

Computing Systems in Engineering

View full text Add to dashboard Cite

E cient data motion is critical for high performance computing on distributed memory architectures. The value of some techniques for e cient data motion is illustrated by identifying generic communication primitives. Further, the e ciency of these primitives is demonstrated on three di erent applications using the nite element method for unstructured grids and sparse solvers with di erent communication requirements. For the applications presented, the techniques advocated reduced the communication times by a factor of between 1.5 { 3.

show abstract

Section: Introductionmentioning

confidence: 99%

Communication primitives for unstructured finite element simulations on data parallel architectures

Mathur¹,

Johnsson²

1992

Computing Systems in Engineering

View full text Add to dashboard Cite

show abstract

A global GMRES/multi-grid scheme for an adaptive Cartesian/quad grid flow solver on distributed memory machines

Wang

1997

13th Computational Fluid Dynamics Conference

View full text Add to dashboard Cite

A global multi-grid/GMRES solution methodology on distributed memory machines is successfully developed in this study. To preserve the effectiveness of the multigrid scheme, the grid partitioning is based on the communication graph of the coarsest grid, so that all levels of the multi-grids are located in the same zone (processor). Each node of the graph is weighted with the total number of the finest grid cells contained in the coarsest grid cell corresponding to the node. This weighted graph is then partitioned with either a Recursive Spectral Bisection (RSB) algorithm or a Recursive Coordinate Bisection (RGB) method to balance the load on the finest grid. With this type of domain decomposition, there is no communication overhead with the prolongation and restriction operations in the multi-grid algorithm. A global GMRES scheme with a multi-grid preconditioner is implemented to drive flows to steady state. The GMRES algorithm is completely parallelizable. Data communication between different zones is provided through a Message Passing Interface (MPI) library, MPICH from Argonne National Laboratory. In addition, a novel communication and computation overlap (CCO) procedure is implemented to further reduce communication time by 10-20%. The performance of the overall algorithm was assessed on the IBM SP2 parallel computer at Maui High Performance Computing Center (MHPCC). Excellent scalability of the overall methodology was demonstrated.

show abstract

Implementation of an Euler/Navier-Stokes finite element algorithm onthe Connection Machine

Cited by 2 publications

References 2 publications

Communication primitives for unstructured finite element simulations on data parallel architectures

Communication primitives for unstructured finite element simulations on data parallel architectures

A global GMRES/multi-grid scheme for an adaptive Cartesian/quad grid flow solver on distributed memory machines

Contact Info

Product

Resources

About