Attending this event?
Back To Schedule
Saturday, January 29 • 11:30am - 12:20pm
Preconditioners to scale Multi-physics Simulations

Sign up or log in to save this to your schedule, view media, leave feedback and see who's attending!

Preconditioners (PCs) are used to improve both, the efficiency and robustness of iterative techniques while solving very large linear systems on a Krylov subspace. However, determining with preconditioner to use with which equations or set of equations on a certain multi-physic simulation requires a combination of knowledge of preconditioning, matrices techniques, types of matrices, Krylov subspaces, iterative methods, among other Linear Algebra's foundation. The present work provides a benchmark of the most popular preconditioners available today, emphasising their respective performance in terms of time to solution of the Finite Element problem, usage of memory, number of iterations, the value of |R| achieved when converged. The performance evaluation is made for the Compute Finite Strain Elastic Stress in 3D, using the University of Cambridge Research Computing Service (CDS3) and the Message Passing Interface (MPI) implementations that allows parallelisation. The benchmark and scaling was done with MOOSE which use the Finite Elements Method and million Degrees of Freedom (DoF). Along with the preconditioners and KSP types, a variety of options were tested to optimise its performance.

Session chairs: Justin Nixon and Michal Ruprich

avatar for Julita Inca

Julita Inca

Member of the GNOME Foundation, GNOME
Computer Science interested. IT Linux and HPC.

Saturday January 29, 2022 11:30am - 12:20pm CET
Session Room 4