Workshop on Solving Large Linear Systems
Martin Bücker (RWTH-SC), Ming Fang (Sun-HPC), Sven Hammarling (NAG), Bruno Lang (RWTH-SC), Arnold Reusken (RWTH-IGPM)
The solution of large systems of linear equations is a fundamental computational task arising from many HPC applications. The availability of the new large Sun compute server offers new possibilities of attacking such problems. The aim of this workshop is to give a survey of methods for solving large linear systems, both sparse and dense, and to give an overview of two high performance libraries specifically tuned for the new Sun SMP-Cluster.
Agenda and Links to the Slides of the Presentations:
13:00 Parallel Direct Methods for Dense Linear Systems (Bruno Lang)
13:45 Parallel Iterative Methods for Sparse Linear Systems (Martin Bücker)
14:30 Preconditioning Techniques for Solving Large Sparse Linear Systems (Arnold Reusken)
15:15 coffee break
15:45 The Sun Scalable Scientific Subroutine Library (S3L) (Ming Fang)
16:45 The NAG High Performance Libraries(Sven Hammarling)
17:45 final discussion
- Martin Buecker is a research scientist at the Computer Science Department of the Aachen University of Technology where he was co-responsible for building up the Institute for Scientific Computing. He received his Diplomas (1993 and 1994) and Ph.D. (1997) degrees in Electrical Engineering and Computer Science from Aachen University of Technology. His research and teaching interests include parallel algorithms, high-performance computing, and automatic differentiation.
- Ming Fang earned her BS and MS in Applied Mathematics from BUAA in China. She obtained her Ph.D in Numerical Analysis at Copenhagen University in 1994. She is a member of the Sun HPC group, developing Sun Scalable Scientific Subroutine Library (S3L) for Sun servers and clusters.
- Sven Hammarling is a Principal Consultant at NAG Ltd (Numerical Algorithms Group) and a Visiting Professor at Cranfield University, RMCS, Shrivenham. He is an author of the BLAS, of LAPACK and of ScaLAPACK.
- Bruno Lang is a senior research assistant at the Computer Science Department at the Aachen University of Technology. He received his diplomas in Computer Science (1989) and Mathematics (1990) and his Ph.D (1991) from Technical University Karlsruhe and his Habilitation (1997) from the Bergische Universität GH Wuppertal. His current activities include teaching and several research projects touching on parallelism, verified computing, and automatic differentiation.
- Arnold Reusken studied Mathematics at the University of Utrecht/NL (1979-1984), where he received his Ph.D (1988). The title of his thesis was "Convergence Analysis of Nonlinear Multigrid Methods". He was assistant professor at the University of Utrecht (1988 -1989) and at the Eindhoven University of Technology (1989 -1997). Since 1997 he is Professor of Numerical Mathematics at the Aachen Technical University (Institut für Geometrie und Praktische Mathematik). His main research interests are numerical treatment of partial differential equations, multigrid methods, and iterative methods for large sparse linear systems.
- Sun Performance Library
The Sun Performance Library offers parallel speedup for a multithreaded process on a single node. With Sun S3L, users can scale performance to multiple processes and multiple nodes. Sun Performance Library is a part of the Sun Forte Developers software package
- Sun Scalable Scientific Subroutine Library (S3L)
Sun S3L provides a variety of functions that are common in scientific and engineering computing, scaling performance to multiple nodes. Sun S3L is a part of the Sun HPC ClusterTools software package.
- The Fortran SMP Library
The 231 routines in NAG's Fortran SMP Library enable developers and programmers to make optimal use of the processing power and shared memory parallelism of their Symmetric Multi-Processor (SMP) systems. The Fortran SMP Library achieves significant performance gains over the hardware vendors' LAPACK implementations.
- The NAG Parallel Library
For anyone involved in true parallel programming, NAG has developed an MPI Parallel Library. This library contains 183 routines that have been specifically developed for use on distributed memory systems and clusters of workstations and PCs.
Dieter an Mey
Phone: +49 241 804377