Adaptive approach to data placement

David K Lowenthal, Gregory R. Andrews

Research output: Chapter in Book/Report/Conference proceedingConference contribution

8 Citations (Scopus)

Abstract

Programming distributed-memory machines requires careful placement of data to balance the computational load among the nodes and minimize excess data movement between the nodes. Most current approaches to data placement require the programmer or compiler to place data initially and then possibly to move it explicitly during a computation. This paper describes a new, adaptive approach. It is implemented in the Adapt system, which takes an initial data placement, efficiently monitors how well it performs, and changes the placement whenever the monitoring indicates that a different placement would perform better. Adapt frees the programmer from having to specify data placements, and it can use run-time information to find better placements than compilers. Moreover, Adapt automatically supports a 'variable block' placement, which is especially useful for applications with nearest-neighbor communication but an imbalanced workload. For applications in which the best data placement varies dynamically, using Adapt can lead to better performance than using any statically determined data placement.

Original languageEnglish (US)
Title of host publicationIEEE Symposium on Parallel and Distributed Processing - Proceedings
PublisherIEEE
Pages349-353
Number of pages5
StatePublished - 1996
EventProceedings of the 1996 10th International Parallel Processing Symposium - Honolulu, HI, USA
Duration: Apr 15 1996Apr 19 1996

Other

OtherProceedings of the 1996 10th International Parallel Processing Symposium
CityHonolulu, HI, USA
Period4/15/964/19/96

Fingerprint

Computer programming
Data storage equipment
Monitoring
Communication

ASJC Scopus subject areas

  • Engineering(all)

Cite this

Lowenthal, D. K., & Andrews, G. R. (1996). Adaptive approach to data placement. In IEEE Symposium on Parallel and Distributed Processing - Proceedings (pp. 349-353). IEEE.

Adaptive approach to data placement. / Lowenthal, David K; Andrews, Gregory R.

IEEE Symposium on Parallel and Distributed Processing - Proceedings. IEEE, 1996. p. 349-353.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Lowenthal, DK & Andrews, GR 1996, Adaptive approach to data placement. in IEEE Symposium on Parallel and Distributed Processing - Proceedings. IEEE, pp. 349-353, Proceedings of the 1996 10th International Parallel Processing Symposium, Honolulu, HI, USA, 4/15/96.
Lowenthal DK, Andrews GR. Adaptive approach to data placement. In IEEE Symposium on Parallel and Distributed Processing - Proceedings. IEEE. 1996. p. 349-353
Lowenthal, David K ; Andrews, Gregory R. / Adaptive approach to data placement. IEEE Symposium on Parallel and Distributed Processing - Proceedings. IEEE, 1996. pp. 349-353
@inproceedings{bb4c8a281aef4e2eb155e353afcd893d,
title = "Adaptive approach to data placement",
abstract = "Programming distributed-memory machines requires careful placement of data to balance the computational load among the nodes and minimize excess data movement between the nodes. Most current approaches to data placement require the programmer or compiler to place data initially and then possibly to move it explicitly during a computation. This paper describes a new, adaptive approach. It is implemented in the Adapt system, which takes an initial data placement, efficiently monitors how well it performs, and changes the placement whenever the monitoring indicates that a different placement would perform better. Adapt frees the programmer from having to specify data placements, and it can use run-time information to find better placements than compilers. Moreover, Adapt automatically supports a 'variable block' placement, which is especially useful for applications with nearest-neighbor communication but an imbalanced workload. For applications in which the best data placement varies dynamically, using Adapt can lead to better performance than using any statically determined data placement.",
author = "Lowenthal, {David K} and Andrews, {Gregory R.}",
year = "1996",
language = "English (US)",
pages = "349--353",
booktitle = "IEEE Symposium on Parallel and Distributed Processing - Proceedings",
publisher = "IEEE",

}

TY - GEN

T1 - Adaptive approach to data placement

AU - Lowenthal, David K

AU - Andrews, Gregory R.

PY - 1996

Y1 - 1996

N2 - Programming distributed-memory machines requires careful placement of data to balance the computational load among the nodes and minimize excess data movement between the nodes. Most current approaches to data placement require the programmer or compiler to place data initially and then possibly to move it explicitly during a computation. This paper describes a new, adaptive approach. It is implemented in the Adapt system, which takes an initial data placement, efficiently monitors how well it performs, and changes the placement whenever the monitoring indicates that a different placement would perform better. Adapt frees the programmer from having to specify data placements, and it can use run-time information to find better placements than compilers. Moreover, Adapt automatically supports a 'variable block' placement, which is especially useful for applications with nearest-neighbor communication but an imbalanced workload. For applications in which the best data placement varies dynamically, using Adapt can lead to better performance than using any statically determined data placement.

AB - Programming distributed-memory machines requires careful placement of data to balance the computational load among the nodes and minimize excess data movement between the nodes. Most current approaches to data placement require the programmer or compiler to place data initially and then possibly to move it explicitly during a computation. This paper describes a new, adaptive approach. It is implemented in the Adapt system, which takes an initial data placement, efficiently monitors how well it performs, and changes the placement whenever the monitoring indicates that a different placement would perform better. Adapt frees the programmer from having to specify data placements, and it can use run-time information to find better placements than compilers. Moreover, Adapt automatically supports a 'variable block' placement, which is especially useful for applications with nearest-neighbor communication but an imbalanced workload. For applications in which the best data placement varies dynamically, using Adapt can lead to better performance than using any statically determined data placement.

UR - http://www.scopus.com/inward/record.url?scp=0029713193&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0029713193&partnerID=8YFLogxK

M3 - Conference contribution

SP - 349

EP - 353

BT - IEEE Symposium on Parallel and Distributed Processing - Proceedings

PB - IEEE

ER -