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

S. Sharma; R. K. Iyer

doi:10.1109/HIPC.1998.738013

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

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

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%.

Original language	English (US)
Title of host publication	Proceedings - 5th International Conference on High Performance Computing, HiPC 1998
Publisher	IEEE Computer Society
Pages	391-398
Number of pages	8
ISBN (Electronic)	0818691948
DOIs	https://doi.org/10.1109/HIPC.1998.738013
State	Published - 1998
Event	5th International Conference on High Performance Computing, HiPC 1998 - Chennai, Madras, India Duration: Dec 17 1998 → Dec 20 1998

Publication series

Name	Proceedings - Symposium on Computer Architecture and High Performance Computing
ISSN (Print)	1550-6533

Other

Other	5th International Conference on High Performance Computing, HiPC 1998
Country/Territory	India
City	Chennai, Madras
Period	12/17/98 → 12/20/98

ASJC Scopus subject areas

Hardware and Architecture
Software

Online availability

10.1109/HIPC.1998.738013

Library availability

Discover UIUC Full Text

Cite this

Sharma, S., & Iyer, R. K. (1998). Measurement-based modeling and analysis methodology for characterizing parallel I/O performance. In Proceedings - 5th International Conference on High Performance Computing, HiPC 1998 (pp. 391-398). (Proceedings - Symposium on Computer Architecture and High Performance Computing). IEEE Computer Society. https://doi.org/10.1109/HIPC.1998.738013

Measurement-based modeling and analysis methodology for characterizing parallel I/O performance. / Sharma, S.; Iyer, R. K.
Proceedings - 5th International Conference on High Performance Computing, HiPC 1998. IEEE Computer Society, 1998. p. 391-398 (Proceedings - Symposium on Computer Architecture and High Performance Computing).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Sharma, S & Iyer, RK 1998, Measurement-based modeling and analysis methodology for characterizing parallel I/O performance. in Proceedings - 5th International Conference on High Performance Computing, HiPC 1998. Proceedings - Symposium on Computer Architecture and High Performance Computing, IEEE Computer Society, pp. 391-398, 5th International Conference on High Performance Computing, HiPC 1998, Chennai, Madras, India, 12/17/98. https://doi.org/10.1109/HIPC.1998.738013

@inproceedings{d5257edfdfa449cc821514d02c422bc1,

title = "Measurement-based modeling and analysis methodology for characterizing parallel I/O performance",

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%.",

author = "S. Sharma and Iyer, {R. K.}",

note = "Publisher Copyright: {\textcopyright} 1998 IEEE.; 5th International Conference on High Performance Computing, HiPC 1998 ; Conference date: 17-12-1998 Through 20-12-1998",

year = "1998",

doi = "10.1109/HIPC.1998.738013",

language = "English (US)",

series = "Proceedings - Symposium on Computer Architecture and High Performance Computing",

publisher = "IEEE Computer Society",

pages = "391--398",

booktitle = "Proceedings - 5th International Conference on High Performance Computing, HiPC 1998",

}

TY - GEN

T1 - Measurement-based modeling and analysis methodology for characterizing parallel I/O performance

AU - Sharma, S.

AU - Iyer, R. K.

PY - 1998

Y1 - 1998

N2 - 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%.

AB - 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%.

UR - http://www.scopus.com/inward/record.url?scp=85045302504&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85045302504&partnerID=8YFLogxK

U2 - 10.1109/HIPC.1998.738013

DO - 10.1109/HIPC.1998.738013

M3 - Conference contribution

AN - SCOPUS:85045302504

T3 - Proceedings - Symposium on Computer Architecture and High Performance Computing

SP - 391

EP - 398

BT - Proceedings - 5th International Conference on High Performance Computing, HiPC 1998

PB - IEEE Computer Society

T2 - 5th International Conference on High Performance Computing, HiPC 1998

Y2 - 17 December 1998 through 20 December 1998

ER -

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

Abstract

Publication series

Other

ASJC Scopus subject areas

Online availability

Library availability

Related links

Fingerprint

Cite this