Wednesday Dec 23 2009, 13:30, at BMB5

Parallel Solution of Large Sparse Linear Systems

Abstract. Recent advances in computer architecture and hardware made petascale platforms possible in which a node consists of many cores. On such platforms, however, one needs novel parallel algorithms that avoid the well-known memory-wall, i.e. design algorithms that minimize memory references compared to minimizing arithmetic operations and simultaneously minimize interprocessor communications. In this talk I present a parallel hybrid sparse linear system solver that is suitable for the solution of large sparse linear systems on parallel computing platforms. The hybrid solver that I will be presenting relies on weighted symmetric and nonsymmetric matrix reordering for bringing the largest elements on or closer to the main diagonal; resulting in a very effective extracted banded preconditioner. Systems involving the extracted banded preconditioner are solved via a member of the recently developed SPIKE family of algorithms. The parallel scalability and the effectiveness of this method is demonstrated by solving large sparse linear systems that arise in various applications, and compared to well known direct and iterative solver packages such as ILUPACK, PARDISO, MUMPS, HYPRE-AMG, and TRILINOS. Some examples of target applications are computational fluid dynamics, electromagnetics and nonlinear optimizations. Finally, I present a highly accurate model for predicting the parallel scalability of the hybrid solver on architectures with more nodes than the platform on which the experiments have been performed.

Biography. Murat Manguoglu was born in Ankara, Turkey, in 1979. In June 2002, he received his bachelor’s degree in Electrical and Electronics Engineering from Middle East Technical University, Ankara, Turkey. He received his master’s degree in Computational Engineering and Science from the University of Utah, Salt Lake City, Utah in May 2004. He joined the Department of Computer Science at Purdue University, West Lafayette, Indiana, as a doctoral student in August 2004. In the summer of 2008, he worked as an intern at Intel Corporation. He completed his doctoral research in May 2009. Currently, he is working as a postdoctoral researcher at Purdue University Computer Science Department.