Scalable Sparse Solvers

Luke Olson, University of Illinois at Urbana-Champaign

Usage Details

The solution to large, sparse linear systems of equations represents a major computational component in a variety of applications. In this project we study the communication demands of algebraic multigrid and the underlying sparse tools that support this method. In particular, we have developed several components to enhance scalability including asynchronous communication, redundant computation, and topology-aware decisions in the algorithm. The Gemini network and full scale of the Blue Waters machine is a unique setting to test our methods and supporting performance models. In addition, as we consider mixed programming models the availability of large cores along with high on-node concurrency is a critical piece of our study. This project focuses on new approaches to sparse matrix-vector multiplication and the scalability of both the algebraic multigrid setup phase as well as scalable multigrid cycling.