In this paper parallel implementation issues concerning computation of a set ofrecurrences on distributed memory parallel systems are discussed. It is a commonproblem when solving partial differential equations on dis...In this paper parallel implementation issues concerning computation of a set ofrecurrences on distributed memory parallel systems are discussed. It is a commonproblem when solving partial differential equations on distributed memory systems,especially in computational fluid dynamics. For example, solution of a set of tri-diagonalsystems of equations or Guauss-Seidel relaxations on finite difference systems all leadto this kind of computation. We emphasize on the idea of doing the computations in apipelined fashion and we show through analysis and numerical examples that by usingproperly chosen parameters, pipelined implementation often yields much better parallelefficiency with respect to other commonly used methods.展开更多
文摘In this paper parallel implementation issues concerning computation of a set ofrecurrences on distributed memory parallel systems are discussed. It is a commonproblem when solving partial differential equations on distributed memory systems,especially in computational fluid dynamics. For example, solution of a set of tri-diagonalsystems of equations or Guauss-Seidel relaxations on finite difference systems all leadto this kind of computation. We emphasize on the idea of doing the computations in apipelined fashion and we show through analysis and numerical examples that by usingproperly chosen parameters, pipelined implementation often yields much better parallelefficiency with respect to other commonly used methods.
