Improving scheduling techniques in heterogeneous systems with dynamic, on-line optimisations

Marcin Bogdański, Peter R. Lewis, Tobias Becker, Xin Yao

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

Abstract

Computational performance increasingly depends on parallelism, and many systems rely on heterogeneous resources such as GPUs and FPGAs to accelerate computationally intensive applications. However, implementations for such heterogeneous systems are often hand-crafted and optimised to one computation scenario, and it can be challenging to maintain high performance when application parameters change. In this paper, we demonstrate that machine learning can help to dynamically choose parameters for task scheduling and load-balancing based on changing characteristics of the incoming workload. We use a financial option pricing application as a case study. We propose a simulation of processing financial tasks on a heterogeneous system with GPUs and FPGAs, and show how dynamic, on-line optimisations could improve such a system. We compare on-line and batch processing algorithms, and we also consider cases with no dynamic optimisations.

LanguageEnglish
Title of host publicationProceedings of the International Conference on Complex, Intelligent and Software Intensive Systems, CISIS 2011
PublisherIEEE
Pages496-501
Number of pages6
ISBN (Electronic)978-0-7695-4373-4
ISBN (Print)978-1-61284-709-2
DOIs
Publication statusPublished - 2011
Event5th International Conference on Complex, Intelligent and Software Intensive Systems - Seoul, Korea, Democratic People's Republic of
Duration: 30 Jun 20112 Jul 2011

Conference

Conference5th International Conference on Complex, Intelligent and Software Intensive Systems
Abbreviated titleCISIS 2011
CountryKorea, Democratic People's Republic of
CitySeoul
Period30/06/112/07/11

Fingerprint

Scheduling
Field programmable gate arrays (FPGA)
Resource allocation
Learning systems
Processing
Costs
Graphics processing unit

Keywords

  • Artificial Neural Network
  • Dynamic Optimisation
  • FPGA
  • Genetic Algorithm
  • GPU
  • Heterogeneous System
  • On-Line Learning
  • Scheduling

Cite this

Bogdański, M., Lewis, P. R., Becker, T., & Yao, X. (2011). Improving scheduling techniques in heterogeneous systems with dynamic, on-line optimisations. In Proceedings of the International Conference on Complex, Intelligent and Software Intensive Systems, CISIS 2011 (pp. 496-501). IEEE. https://doi.org/10.1109/CISIS.2011.81
Bogdański, Marcin ; Lewis, Peter R. ; Becker, Tobias ; Yao, Xin. / Improving scheduling techniques in heterogeneous systems with dynamic, on-line optimisations. Proceedings of the International Conference on Complex, Intelligent and Software Intensive Systems, CISIS 2011. IEEE, 2011. pp. 496-501
@inproceedings{e7421d5512254bccb782abb5ac87918e,
title = "Improving scheduling techniques in heterogeneous systems with dynamic, on-line optimisations",
abstract = "Computational performance increasingly depends on parallelism, and many systems rely on heterogeneous resources such as GPUs and FPGAs to accelerate computationally intensive applications. However, implementations for such heterogeneous systems are often hand-crafted and optimised to one computation scenario, and it can be challenging to maintain high performance when application parameters change. In this paper, we demonstrate that machine learning can help to dynamically choose parameters for task scheduling and load-balancing based on changing characteristics of the incoming workload. We use a financial option pricing application as a case study. We propose a simulation of processing financial tasks on a heterogeneous system with GPUs and FPGAs, and show how dynamic, on-line optimisations could improve such a system. We compare on-line and batch processing algorithms, and we also consider cases with no dynamic optimisations.",
keywords = "Artificial Neural Network, Dynamic Optimisation, FPGA, Genetic Algorithm, GPU, Heterogeneous System, On-Line Learning, Scheduling",
author = "Marcin Bogdański and Lewis, {Peter R.} and Tobias Becker and Xin Yao",
year = "2011",
doi = "10.1109/CISIS.2011.81",
language = "English",
isbn = "978-1-61284-709-2",
pages = "496--501",
booktitle = "Proceedings of the International Conference on Complex, Intelligent and Software Intensive Systems, CISIS 2011",
publisher = "IEEE",
address = "United States",

}

Bogdański, M, Lewis, PR, Becker, T & Yao, X 2011, Improving scheduling techniques in heterogeneous systems with dynamic, on-line optimisations. in Proceedings of the International Conference on Complex, Intelligent and Software Intensive Systems, CISIS 2011. IEEE, pp. 496-501, 5th International Conference on Complex, Intelligent and Software Intensive Systems, Seoul, Korea, Democratic People's Republic of, 30/06/11. https://doi.org/10.1109/CISIS.2011.81

Improving scheduling techniques in heterogeneous systems with dynamic, on-line optimisations. / Bogdański, Marcin; Lewis, Peter R.; Becker, Tobias; Yao, Xin.

Proceedings of the International Conference on Complex, Intelligent and Software Intensive Systems, CISIS 2011. IEEE, 2011. p. 496-501.

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

TY - GEN

T1 - Improving scheduling techniques in heterogeneous systems with dynamic, on-line optimisations

AU - Bogdański, Marcin

AU - Lewis, Peter R.

AU - Becker, Tobias

AU - Yao, Xin

PY - 2011

Y1 - 2011

N2 - Computational performance increasingly depends on parallelism, and many systems rely on heterogeneous resources such as GPUs and FPGAs to accelerate computationally intensive applications. However, implementations for such heterogeneous systems are often hand-crafted and optimised to one computation scenario, and it can be challenging to maintain high performance when application parameters change. In this paper, we demonstrate that machine learning can help to dynamically choose parameters for task scheduling and load-balancing based on changing characteristics of the incoming workload. We use a financial option pricing application as a case study. We propose a simulation of processing financial tasks on a heterogeneous system with GPUs and FPGAs, and show how dynamic, on-line optimisations could improve such a system. We compare on-line and batch processing algorithms, and we also consider cases with no dynamic optimisations.

AB - Computational performance increasingly depends on parallelism, and many systems rely on heterogeneous resources such as GPUs and FPGAs to accelerate computationally intensive applications. However, implementations for such heterogeneous systems are often hand-crafted and optimised to one computation scenario, and it can be challenging to maintain high performance when application parameters change. In this paper, we demonstrate that machine learning can help to dynamically choose parameters for task scheduling and load-balancing based on changing characteristics of the incoming workload. We use a financial option pricing application as a case study. We propose a simulation of processing financial tasks on a heterogeneous system with GPUs and FPGAs, and show how dynamic, on-line optimisations could improve such a system. We compare on-line and batch processing algorithms, and we also consider cases with no dynamic optimisations.

KW - Artificial Neural Network

KW - Dynamic Optimisation

KW - FPGA

KW - Genetic Algorithm

KW - GPU

KW - Heterogeneous System

KW - On-Line Learning

KW - Scheduling

UR - http://www.scopus.com/inward/record.url?scp=80052759109&partnerID=8YFLogxK

UR - http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=5989059

U2 - 10.1109/CISIS.2011.81

DO - 10.1109/CISIS.2011.81

M3 - Conference contribution

SN - 978-1-61284-709-2

SP - 496

EP - 501

BT - Proceedings of the International Conference on Complex, Intelligent and Software Intensive Systems, CISIS 2011

PB - IEEE

ER -

Bogdański M, Lewis PR, Becker T, Yao X. Improving scheduling techniques in heterogeneous systems with dynamic, on-line optimisations. In Proceedings of the International Conference on Complex, Intelligent and Software Intensive Systems, CISIS 2011. IEEE. 2011. p. 496-501 https://doi.org/10.1109/CISIS.2011.81