The concurrent solution of large linear systems of equations on processor meshes of fixed size has been considered taking into account the basic requirements of computational efficiency and stability. A linear rotation based algorithm has been studied as a robust solution of linear problems well suited to be implemented on parallel architectures because of its regular computational structure. The exploitation of the intrinsic parallelism of the algorithm and the application of partitioning techniques have addressed a locally parallel globally sequential parallelization strategy. The method has been experimentally evaluated on different sized arrays of transputers, demonstrating that high computational efficiency can be achieved.
A Linear Rotation Based Solution of Large Systems on a Transputer Array
CHIRICO, MARCO;DI ZITTI, ERMANNO;BISIO, GIACOMO
1995-01-01
Abstract
The concurrent solution of large linear systems of equations on processor meshes of fixed size has been considered taking into account the basic requirements of computational efficiency and stability. A linear rotation based algorithm has been studied as a robust solution of linear problems well suited to be implemented on parallel architectures because of its regular computational structure. The exploitation of the intrinsic parallelism of the algorithm and the application of partitioning techniques have addressed a locally parallel globally sequential parallelization strategy. The method has been experimentally evaluated on different sized arrays of transputers, demonstrating that high computational efficiency can be achieved.
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
