Conference publications

Abstracts

XX conference

Comparative analysis of some parallel algorithms for solving sparse systems of linear algebraic equations in application to numerical solution of the Poisson equation

Nikolsky I.M.

Moscow, Vorobyovi mountains, GSP-2, CMC faculty

It is becoming more and more popular to use supercomputers for numerical solution of PDEs. This inevitably raises the question of choice of the best method for solving sparse systems of linear equations arising as a result of PDE discretization.

This work presents some results of comparative analysis of effectiveness of some parallel algorithms for solving sparse linear systems, included in software library Hypre. List of methods includes Krylov subspaces methods and some multigrid varieties (including algebraic multigrid). Boundary problems for Poisson equation in different areas were used as model problems. All computations were performed on BlueGene/P machine owned by CMC faculty of MSU.

Obtained results enable estimation of thefinest mesh that can be used for solving model problem in a fixed time on a fixed number of CPUs. In the talk there also be discussed numerical solution verification and frequency analysis of iteration methods convergence

• view as PDF (in Russian) (PDF)