Goal: This project proposes to investigate, produce, and maintain a methodology and its software implementation that leverage emerging heterogeneous hardware architectures to solve billion-unknowns linear systems in a robust, scalable, and efficient fashion. The two classes of problems targeted under this project are banded dense and sparse general linear systems. Preliminary results suggest that the adopted methodology displays a good strong-scaling attribute and its early implementation, called SaP (for Splitting-and-Partitioning), is one order of magnitude faster than competitive software solutions.
Motivation: The task of solving a linear system is one of the most ubiquitous ingredients in the numerical solution of Applied Mathematics problems. It is relied upon for the implicit integration of Ordinary Differential Equation (ODE) and Differential Algebraic Equation (DAE) problems, in the numerical solution of Partial Differential Equation (PDE) problems, in interior point optimization methods, in least squares approximations, in solving eigenvalue problems, and in data analysis. In fact, the vast majority of nonlinear problems in Scientific Computing are solved iteratively by drawing on local linearizations of nonlinear operators and the solution of linear systems. Recent advances in (a) hardware architecture; i.e., the emergence of General Purpose Graphics Processing Unit (GP-GPU) cards, and (b) scalable solution algorithms, provide an opportunity to develop a new class of parallel algorithms that can robustly and efficiently solve very large linear systems of equations.