A Parallel Crout Algorithm based on TBB

Author

Zhang, Liyan ; Sun, Yan ; Ma, Jian

Author_Institution

Sch. of Transp. Eng., Tongji Univ., Shanghai, China

fYear

2011

fDate

15-17 July 2011

Firstpage

239

Lastpage

242

Abstract

The paper presents a novel Parallel Crout Algorithm (PCA) based on multi-core computer with Threading Building Blocks (TBB). TBB offers a rich and complete approach to express parallelism in a C++ program. PCA is decomposed into three-tier: data decomposition parallelism, task processing parallelism and data composition parallelism and it can improve the efficiency of solving linear systems. Compared with Sequential Crout Algorithm (SCA), PCA has advantages of high efficiency, cross-platform and scalability. SCA and PCA, which is based on TBB, are implemented with C++. The validities of both methods are verified by different scale of matrix. In order to improve decomposition rate, the paper optimizes the parameters of PCA. Experiments show that, compared with SCA, PCA can reached a faster solution speed and a higher efficiency and it takes full advantage of Symmetrical Multi-Processing computer.

Keywords

C++ language; multiprocessing systems; parallel algorithms; C++ program; TBB; data composition parallelism; data decomposition parallelism; multicore computer; parallel crout algorithm; sequential crout algorithm; symmetrical multiprocessing computer; task processing parallelism; threading building blocks; Crout Decomposition; Matrix Decomposition; Multi-thread; Threading Building Blocks (TBB);

fLanguage

English

Publisher

ieee

Conference_Titel

Software Engineering and Service Science (ICSESS), 2011 IEEE 2nd International Conference on

Conference_Location

Beijing

Print_ISBN

978-1-4244-9699-0

Type

conf

DOI

10.1109/ICSESS.2011.5982298

Filename

5982298