Parallel Solution of a General Sparse Linear System Using pARMS

Yousef Saad

Department of Computer Science and Engineering, University of Minnesota, 200 Union Street S.E., Minneapolis, MN 55455

Masha Sosonkina

Department of Computer Science, University of Minnesota - Duluth, 320 Heller Hall, 1114 Kirby Dr., Duluth, MN 55812


Abstract

The parallel Algebraic Recursive Multilevel Solver (pARMS) is a recently developed suite of parallel iterative methods for the solution of large sparse linear systems. The package contains a vast array of parallel preconditioners based on domain decomposition ideas, as well as algebraic multilevel concepts. A single framework, which encompasses all the preconditioner construction, has been made possible due to the "distributed sparse linear system" design principle and the adoption of a two-level interface nodes approach. The pARMS package has been tested on several computing platforms, such as a Linux cluster, IBM SP, and SGI Origin. After presenting the design rationale, the talk will report on the current status of pARMS, and show the results of some recent experiments.