Measurement-based modeling and analysis methodology for characterizing parallel I/O performance

Author

Sharma, S. ; Iyer, R.K.

Author_Institution

Coordinated Sci. Lab., Illinois Univ., Urbana, IL, USA

fYear

1998

fDate

17-20 Dec 1998

Firstpage

391

Lastpage

398

Abstract

A parallel I/O characterization methodology that consists of a hierarchical modeling and measurement analysis environment for investigating I/O performance is presented. The methodology is illustrated via a case study of a video server workload running under the parallel I/O file system (PIOFS) of IBM SP/2. The measurements demonstrate that for video server and read-intensive workloads, spreading parallel files across all eight I/O servers improves a client´s bandwidth performance by 36-52%. With eight clients, the per-client bandwidth performance increases by only 15%-23%. PIOFS-based default file striping results in degradation of bandwidth performance by as much as 25%

Keywords

client-server systems; input-output programs; parallel processing; performance evaluation; resource allocation; video servers; I/O servers; IBM SP/2; PIOFS-based default file striping; client bandwidth performance; hierarchical measurement analysis environment; hierarchical modeling environment; measurement-based analysis methodology; measurement-based modeling methodology; parallel I/O file system; parallel I/O performance characterization; parallel file spreading; read-intensive workloads; video server workload; Bandwidth; Degradation; File servers; File systems; Laboratories; NASA; Network servers; Performance analysis; Performance evaluation; Software systems;

fLanguage

English

Publisher

ieee

Conference_Titel

High Performance Computing, 1998. HIPC '98. 5th International Conference On

Conference_Location

Madras

Print_ISBN

0-8186-9194-8

Type

conf

DOI

10.1109/HIPC.1998.738013

Filename

738013