Parallel Sparse Linear System and Eigenvalue Problem Solvers: From Multicore to Petascale Computing.

摘要

Sparse matrix computations arise in numerous computational science and engineering computations as well as in network analysis and databased simulations. On parallel computing platforms, however, sparse matrix computations represent a major impediment to realizing high performance. Our project aims at designing and implementing solvers for: (i) large sparse linear systems, and (ii) large sparse symmetric eigenvalue problems that achieve high performance on a single multicore node and clusters of many multicore nodes. Further, we demonstrate both the superior robustness and parallel scalability of our solvers compared to other publicly available parallel solvers for these two fundamental problems.