We present a solver for a class of sparse linear systems that we call quasi block diagonal. The solver combines multi-processors and multi-threaded parallelisms using MPI and OpenMP to implement preconditioned Jacobi. Specific formats for sparse matrices are exploited in order to reduce memory storage requirements. Our experiments show that communication costs are negligible, so as that speed-up and efficiency with respect to the sequential implementation are very high. Our hybrid implementation is tested on a cluster and compared to Intel MKL PARDISO linear solver.
Hybrid Solver for Quasi Block Diagonal Linear Systems / Arrigoni, Viviana; Massini, Annalisa. - 12043:(2020), pp. 129-140. (Intervento presentato al convegno Parallel Processing and Applied Mathematics tenutosi a Bialystok, Poland, September 8-11, 2019) [10.1007/978-3-030-43229-4_12].
Hybrid Solver for Quasi Block Diagonal Linear Systems
Arrigoni, Viviana;Massini, Annalisa
2020
Abstract
We present a solver for a class of sparse linear systems that we call quasi block diagonal. The solver combines multi-processors and multi-threaded parallelisms using MPI and OpenMP to implement preconditioned Jacobi. Specific formats for sparse matrices are exploited in order to reduce memory storage requirements. Our experiments show that communication costs are negligible, so as that speed-up and efficiency with respect to the sequential implementation are very high. Our hybrid implementation is tested on a cluster and compared to Intel MKL PARDISO linear solver.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
