Autonomic runtime system for large scale parallel and distributed applications

Jingmei Yang, Huoping Chen, Byoung Uk Kim, Salim A Hariri, Manish Parashar

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

Abstract

The development of efficient parallel algorithms for large scale wildfire simulations is a challenging research problem because the factors that determine wildfire behavior are complex; they include fuel characteristics and configurations, chemical reactions, balances between different modes of heat transfer, topography, and fire/atmosphere interactions. These factors make static parallel algorithms inefficient, especially when large number of processors are used because we cannot predict accurately the propagation of the fire and its computational requirements at runtime. In this paper, we present an Autonomic Runtime Manager (ARM) to dynamically exploit the physics properties of the fire simulation and use them as the basis of our self-optimization algorithm. At each step of the wildfire simulation, the ARM decomposes the computational domain into several natural regions (e.g., burning, unburned, burned) where each region has the same temporal and special characteristics. The number of burning, unburned and burned cells determines the current state of the fire simulation and can then be used to accurately predict the computational power required for each region. By regularly monitoring the state of the simulation and analyzing it, and use that to drive the runtime optimization, we can achieve significant performance gains because we can efficiently balance the computational load on each processor. Our experimental results show that the performance of the fire simulation has been improved by 45% when compared with a static portioning algorithm that does not take into considerations the state of the computations.

Original languageEnglish (US)
Title of host publicationLecture Notes in Computer Science
EditorsJ.-P. Banatre, P. Fradet, J.-L. Giavitto, O. Michel
Pages297-311
Number of pages15
Volume3566
StatePublished - 2005
EventInternational Workshop on Unconventional Programming Paradigms, UPP 2004 - Le Mont Saint Michel, France
Duration: Sep 15 2004Sep 17 2004

Other

OtherInternational Workshop on Unconventional Programming Paradigms, UPP 2004
CountryFrance
CityLe Mont Saint Michel
Period9/15/049/17/04

Fingerprint

Fires
Parallel algorithms
Managers
Topography
Chemical reactions
Physics
Heat transfer
Monitoring

ASJC Scopus subject areas

  • Computer Science (miscellaneous)

Cite this

Yang, J., Chen, H., Kim, B. U., Hariri, S. A., & Parashar, M. (2005). Autonomic runtime system for large scale parallel and distributed applications. In J-P. Banatre, P. Fradet, J-L. Giavitto, & O. Michel (Eds.), Lecture Notes in Computer Science (Vol. 3566, pp. 297-311)

Autonomic runtime system for large scale parallel and distributed applications. / Yang, Jingmei; Chen, Huoping; Kim, Byoung Uk; Hariri, Salim A; Parashar, Manish.

Lecture Notes in Computer Science. ed. / J.-P. Banatre; P. Fradet; J.-L. Giavitto; O. Michel. Vol. 3566 2005. p. 297-311.

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

Yang, J, Chen, H, Kim, BU, Hariri, SA & Parashar, M 2005, Autonomic runtime system for large scale parallel and distributed applications. in J-P Banatre, P Fradet, J-L Giavitto & O Michel (eds), Lecture Notes in Computer Science. vol. 3566, pp. 297-311, International Workshop on Unconventional Programming Paradigms, UPP 2004, Le Mont Saint Michel, France, 9/15/04.
Yang J, Chen H, Kim BU, Hariri SA, Parashar M. Autonomic runtime system for large scale parallel and distributed applications. In Banatre J-P, Fradet P, Giavitto J-L, Michel O, editors, Lecture Notes in Computer Science. Vol. 3566. 2005. p. 297-311
Yang, Jingmei ; Chen, Huoping ; Kim, Byoung Uk ; Hariri, Salim A ; Parashar, Manish. / Autonomic runtime system for large scale parallel and distributed applications. Lecture Notes in Computer Science. editor / J.-P. Banatre ; P. Fradet ; J.-L. Giavitto ; O. Michel. Vol. 3566 2005. pp. 297-311
@inproceedings{6949ced91e8f43718c97a6ff2a52a56a,
title = "Autonomic runtime system for large scale parallel and distributed applications",
abstract = "The development of efficient parallel algorithms for large scale wildfire simulations is a challenging research problem because the factors that determine wildfire behavior are complex; they include fuel characteristics and configurations, chemical reactions, balances between different modes of heat transfer, topography, and fire/atmosphere interactions. These factors make static parallel algorithms inefficient, especially when large number of processors are used because we cannot predict accurately the propagation of the fire and its computational requirements at runtime. In this paper, we present an Autonomic Runtime Manager (ARM) to dynamically exploit the physics properties of the fire simulation and use them as the basis of our self-optimization algorithm. At each step of the wildfire simulation, the ARM decomposes the computational domain into several natural regions (e.g., burning, unburned, burned) where each region has the same temporal and special characteristics. The number of burning, unburned and burned cells determines the current state of the fire simulation and can then be used to accurately predict the computational power required for each region. By regularly monitoring the state of the simulation and analyzing it, and use that to drive the runtime optimization, we can achieve significant performance gains because we can efficiently balance the computational load on each processor. Our experimental results show that the performance of the fire simulation has been improved by 45{\%} when compared with a static portioning algorithm that does not take into considerations the state of the computations.",
author = "Jingmei Yang and Huoping Chen and Kim, {Byoung Uk} and Hariri, {Salim A} and Manish Parashar",
year = "2005",
language = "English (US)",
volume = "3566",
pages = "297--311",
editor = "J.-P. Banatre and P. Fradet and J.-L. Giavitto and O. Michel",
booktitle = "Lecture Notes in Computer Science",

}

TY - GEN

T1 - Autonomic runtime system for large scale parallel and distributed applications

AU - Yang, Jingmei

AU - Chen, Huoping

AU - Kim, Byoung Uk

AU - Hariri, Salim A

AU - Parashar, Manish

PY - 2005

Y1 - 2005

N2 - The development of efficient parallel algorithms for large scale wildfire simulations is a challenging research problem because the factors that determine wildfire behavior are complex; they include fuel characteristics and configurations, chemical reactions, balances between different modes of heat transfer, topography, and fire/atmosphere interactions. These factors make static parallel algorithms inefficient, especially when large number of processors are used because we cannot predict accurately the propagation of the fire and its computational requirements at runtime. In this paper, we present an Autonomic Runtime Manager (ARM) to dynamically exploit the physics properties of the fire simulation and use them as the basis of our self-optimization algorithm. At each step of the wildfire simulation, the ARM decomposes the computational domain into several natural regions (e.g., burning, unburned, burned) where each region has the same temporal and special characteristics. The number of burning, unburned and burned cells determines the current state of the fire simulation and can then be used to accurately predict the computational power required for each region. By regularly monitoring the state of the simulation and analyzing it, and use that to drive the runtime optimization, we can achieve significant performance gains because we can efficiently balance the computational load on each processor. Our experimental results show that the performance of the fire simulation has been improved by 45% when compared with a static portioning algorithm that does not take into considerations the state of the computations.

AB - The development of efficient parallel algorithms for large scale wildfire simulations is a challenging research problem because the factors that determine wildfire behavior are complex; they include fuel characteristics and configurations, chemical reactions, balances between different modes of heat transfer, topography, and fire/atmosphere interactions. These factors make static parallel algorithms inefficient, especially when large number of processors are used because we cannot predict accurately the propagation of the fire and its computational requirements at runtime. In this paper, we present an Autonomic Runtime Manager (ARM) to dynamically exploit the physics properties of the fire simulation and use them as the basis of our self-optimization algorithm. At each step of the wildfire simulation, the ARM decomposes the computational domain into several natural regions (e.g., burning, unburned, burned) where each region has the same temporal and special characteristics. The number of burning, unburned and burned cells determines the current state of the fire simulation and can then be used to accurately predict the computational power required for each region. By regularly monitoring the state of the simulation and analyzing it, and use that to drive the runtime optimization, we can achieve significant performance gains because we can efficiently balance the computational load on each processor. Our experimental results show that the performance of the fire simulation has been improved by 45% when compared with a static portioning algorithm that does not take into considerations the state of the computations.

UR - http://www.scopus.com/inward/record.url?scp=26444496555&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=26444496555&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:26444496555

VL - 3566

SP - 297

EP - 311

BT - Lecture Notes in Computer Science

A2 - Banatre, J.-P.

A2 - Fradet, P.

A2 - Giavitto, J.-L.

A2 - Michel, O.

ER -