Hybrid programming implementation and analysis of CG method in high-performance cluster

Author

Shi, Lijuan ; Shi, Weifeng

Author_Institution

Sch. of Inf. Sci. & Technol., Northwest Univ., Xi´´an, China

Volume

5

fYear

2010

fDate

25-27 June 2010

Abstract

CG Method is the most prominent iterative method for solving sparse systems of linear equations. Implementing the CG method in high-performance cluster has practical significance in many fields such as science and engineering. The advantages of using hybrid MPI+OpenMP programming in high-performance cluster are full use of cluster´s features between the nodes with a distributed storage and within nodes shared storage. Coarse-grained message passing communication is implemented by MPI between the nodes. Fine-grained shared-memory communication is implemented by OpenMP within the nodes, which reduces the running time of parallel algorithm.

Keywords

conjugate gradient methods; iterative methods; message passing; parallel algorithms; shared memory systems; coarse-grained message passing communication; conjugate gradient method; distributed storage; fine-grained shared-memory communication; high-performance cluster; hybrid MPI+OpenMP programming; hybrid programming implementation; iterative method; linear equation; nodes shared storage; parallel algorithm; sparse system; Blades; Character generation; Concurrent computing; Equations; Information science; Libraries; Linear programming; Message passing; Parallel programming; Yarn; CG method; MPI+OpenMP; hybrid programming; multi-core cluster;

fLanguage

English

Publisher

ieee

Conference_Titel

Computer Design and Applications (ICCDA), 2010 International Conference on

Conference_Location

Qinhuangdao

Print_ISBN

978-1-4244-7164-5

Electronic_ISBN

978-1-4244-7164-5

Type

conf

DOI

10.1109/ICCDA.2010.5541197

Filename

5541197