Distributed task processing performance on a NUMA shared memory multiprocessor

Author

Zhang, Xiaodong ; Srinivasan, Pattabiraman

Author_Institution

Div. of Math. & Comput. Sci., Texas Univ., San Antonio, TX, USA

fYear

1990

fDate

9-13 Dec 1990

Firstpage

786

Lastpage

789

Abstract

The effects of distributed task processing on a NUMA (nonuniform memory access) shared memory multiprocessor are studied and examined. Several analytical models are given for describing the behavior of different scheduling strategies. The performance of pre-scheduling and self-scheduling on the BBN GP1000, a NUMA shared multiprocessor is measured through several application programs. The experimental results are consistent with the analytical models. The results may be used to determine an optimal scheduling strategy for effective use of a NUMA shared memory multiprocessor in developing an efficient parallel programming environment and running application programs

Keywords

distributed processing; multiprocessing systems; parallel processing; performance evaluation; resource allocation; scheduling; BBN GP1000; NUMA shared memory multiprocessor; NUMA shared multiprocessor; distributed task processing; optimal scheduling strategy; parallel programming environment; scheduling strategies; Analytical models; Computer science; Mathematics; Memory architecture; Memory management; Message passing; Optimal scheduling; Programming environments; Runtime; Switches;

fLanguage

English

Publisher

ieee

Conference_Titel

Parallel and Distributed Processing, 1990. Proceedings of the Second IEEE Symposium on

Conference_Location

Dallas, TX

Print_ISBN

0-8186-2087-0

Type

conf

DOI

10.1109/SPDP.1990.143645

Filename

143645