Optimal-processor-set speedup of SMP cluster

Author

He, Jiahua ; Chen, Guoliang ; Shan, Jiulong

Author_Institution

Dept. of Comput. Sci. & Technol., Univ. of Sci. & Technol. of China, Anhui, China

fYear

2002

fDate

23-25 Oct. 2002

Firstpage

329

Lastpage

332

Abstract

The SMP cluster becomes more and more important in high performance computing and speedup is the main performance criterion of parallel and distributed algorithms. But the traditional definition of speedup is not fit for the SMP cluster architecture. How can we measure speedup of the SMP cluster? We analyze the nonequivalence of n-processor sets in the SMP cluster based on the introduction to traditional and node-oriented speedup definitions, propose a corresponding speedup definition, optimal-processor-set speedup, in both theoretical and practical views, and illustrate its benefits by an example executing on two SMP clusters.

Keywords

parallel algorithms; parallel architectures; shared memory systems; software performance evaluation; workstation clusters; PC cluster; SMP cluster; distributed algorithms; high performance computing; node-oriented speedup; optimal-processor-set speedup; parallel algorithms; parallel architecture; performance evaluation; Computer architecture; Computer science; Distributed algorithms; Distributed computing; Helium; High performance computing; Partial response channels; Scalability; Usability; Velocity measurement;

fLanguage

English

Publisher

ieee

Conference_Titel

Algorithms and Architectures for Parallel Processing, 2002. Proceedings. Fifth International Conference on

Conference_Location

Beijing, China

Print_ISBN

0-7695-1512-6

Type

conf

DOI

10.1109/ICAPP.2002.1173596

Filename

1173596