Abstract
A performance evaluation framework for the solution schemes for sparse linear systems is proposed. The framework systematically constructs a performance database that provides a visual diagram of solution algorithms’ performance and characteristics to represent the relationship between the solution algorithms and solution problems. In addition, the database model is best used with software engineering techniques to facilitate automatic tuning of sparse linear solvers. This approach resembles the techniques used in quality control. Two types of cases using this approach are presented. One involves knowledge discovery in a database and reveals that preconditioning is more effective than the choice of solver for obtaining rapid convergence of iterative solutions. The other case is an improvement in quality related to numerical solving processes.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
- 2.
Although the matrix data analysis method for quality control is sometimes taken to refer specifically to “principal component analysis,” it is interpreted in a broader sense in this study.
- 3.
This diagram only is the result of not designating “-saamg_ unsymtrue” in the Lis SA-AMG parameter. In the other diagrams, “-saamg _unsymtrue” is designated. Due to the existence of this option, there is no major change in the position of cells which become gray color, and there is almost no effect on the discussion in this section.
- 4.
After the lis-1.2.0 version, the problem of these two solvers was modified.
References
Barrett R et al (1994) Templates for the solution of linear systems: Building blocks for iterative methods. SIAM, PA
Itoh S (2009) Systematic performance evaluation for numerical algorithms of linear equations and its knowledge discovery. SIAM CSE09, Mar 2009
Itoh S (2009) Software tools to assist in automatic tuning. In: 4th international workshop on automatic performance tuning (iWAPT2009), The University of Tokyo, Oct 2009
Itoh S, Sugihara M (2008) A quality management approach for systematic performance evaluations of numerical solving process of linear equations. In: Proceedings of Japan SIAM annual meeting, Kashiwa, Chiba, Sept 2008 (in Japanese)
Moriguti S (ed) (1990) Software quality control guidebook. Japanese Standards Association (in Japanese)
Soin SS (1998) Total quality essentials, 2nd edn. McGraw-Hill, NY
Itoh S, Kotakemori H, Hasegawa H (2007) Development of evaluation system for numerical algorithms to solve linear equations. In: Wenbin L et al (eds) Recent progress in scientific computing. Science Press, Beijing, pp 231–241
SALSA. http://icl.cs.utk.edu/salsa/
GRID-TLSE. http://gridtlse.org/
Li XS, Marques OA et al (2006) EigAdept – The expert eigensolver toolbox. In: Proceedings of workshop on state-of-the-art in scientific and parallel computing (PARA’06), 191, Umeå, Sweden, June 2006
Gould NI, Scott JA (2007) A numerical evaluation of sparse direct solvers for the solution of large sparse symmetric linear systems of equations. ACM Trans Math Software 33(2):1–32
Bhowmick S, Raghavan P, Teranishi K (2002) A combinatorial scheme for developing efficient composite solvers. Lecture Notes in Computer Science, vol 2330, Computational Science-ICCS 2002. Springer, Berlin, pp 325–334
Matrix Market. http://math.nist.gov/MatrixMarket/
Moriguti S (ed) (2003) Statistical methods. Japanese Standards Association (in Japanese)
Itoh S, Sugihara M (2009) A study of a relation between preconditioned Krylov subspace methods and their convergence criterions to solve linear equations. IPSJ SIG Technical Report, vol 2009-HPC-121, no 5, p 8 (in Japanese)
Itoh S, Sugihara M (2009) Systematic property analysis on preconditioned systems of Krylov subspace methods to solve linear equations. In: Proceedings of Japan SIAM annual meeting, Osaka, Japan, Sept 2009 (in Japanese)
Acknowledgments
The first author (S.I.) sincerely expresses his gratitude to the members of the Japanese Automatic Tuning Research Group for their valuable discussions with him. He also extends his gratitude to Dr. Hisashi Kotakemori of TCAD International, Inc. for his valuable advice regarding programming with the Lis library. This work is partially supported by Grant-in-Aid for Scientific Research (B) “Development of the Framework to Support Large-scale Numerical Simulation on Multi-platform,” No.21300017; Grant-in-Aid for Scientific Research (B) “Development of Auto-tuning Specification Language Towards Manycore and Massively Parallel Processing Era,” No. 21300007; and Grant-in-Aid for Scientific Research (B) “A study on Autotuning enhanced by Hierarchical Algorithm Selections,” No. 20300007 of MEXT Japan; and the “Next-Generation Integrated Simulation of Living Matter” of RIKEN.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2011 Springer New York
About this chapter
Cite this chapter
Itoh, S., Sugihara, M. (2011). Systematic Performance Evaluation of Linear Solvers Using Quality Control Techniques. In: Naono, K., Teranishi, K., Cavazos, J., Suda, R. (eds) Software Automatic Tuning. Springer, New York, NY. https://doi.org/10.1007/978-1-4419-6935-4_9
Download citation
DOI: https://doi.org/10.1007/978-1-4419-6935-4_9
Published:
Publisher Name: Springer, New York, NY
Print ISBN: 978-1-4419-6934-7
Online ISBN: 978-1-4419-6935-4
eBook Packages: EngineeringEngineering (R0)