Mapping of large scale fluid power system simulations on a distributed memory parallel computer using genetic algorithms

K Pollmeier, C R Burrows, K A Edge

Research output: Contribution to journalArticle

2 Citations (Scopus)

Abstract

Parallel simulation of fluid power systems using the transmission-line modeling method offers the benefit of increased speed of execution, but requires the system model to be partitioned on to individual processors [Burton, 1994]. In this paper we address the automatic placement of component models on processors of a distributed memory parallel machine. A genetic algorithm is used to map the different processes onto several processors. The objective is to minimise the total processing time in order to achieve real time performance. For fine grained computation problems the communication time cannot be neglected, i.e. we consider the computation and the communication time of each task. This mapping problem is a combinatorial optimization problem which can be reduced to the graph partitioning problem which is shown to be NP-complete. Combining genetic algorithms with heuristics leads to optimal or very good sub-optimal solutions. A hydraulic example circuit is partitioned in four and eight subsystems, respectively, and the simulation is implemented on a Transputer based platform. Using eight processors a speed up of 3.7 was achieved
Original languageEnglish
Pages (from-to)83-91
Number of pages9
JournalAmerican Society of Mechanical Engineers, The Fluid Power and Systems Technology Division (Publication) FPST
Volume3
Publication statusPublished - 1996

Fingerprint

Computer systems
Genetic algorithms
Transputers
Data storage equipment
Fluids
Communication
Combinatorial optimization
Electric lines
Hydraulics
Networks (circuits)
Processing

Cite this

@article{ef8253dd68ef4e9189c7505a00ed8e32,
title = "Mapping of large scale fluid power system simulations on a distributed memory parallel computer using genetic algorithms",
abstract = "Parallel simulation of fluid power systems using the transmission-line modeling method offers the benefit of increased speed of execution, but requires the system model to be partitioned on to individual processors [Burton, 1994]. In this paper we address the automatic placement of component models on processors of a distributed memory parallel machine. A genetic algorithm is used to map the different processes onto several processors. The objective is to minimise the total processing time in order to achieve real time performance. For fine grained computation problems the communication time cannot be neglected, i.e. we consider the computation and the communication time of each task. This mapping problem is a combinatorial optimization problem which can be reduced to the graph partitioning problem which is shown to be NP-complete. Combining genetic algorithms with heuristics leads to optimal or very good sub-optimal solutions. A hydraulic example circuit is partitioned in four and eight subsystems, respectively, and the simulation is implemented on a Transputer based platform. Using eight processors a speed up of 3.7 was achieved",
author = "K Pollmeier and Burrows, {C R} and Edge, {K A}",
year = "1996",
language = "English",
volume = "3",
pages = "83--91",
journal = "American Society of Mechanical Engineers, The Fluid Power and Systems Technology Division (Publication) FPST",

}

TY - JOUR

T1 - Mapping of large scale fluid power system simulations on a distributed memory parallel computer using genetic algorithms

AU - Pollmeier, K

AU - Burrows, C R

AU - Edge, K A

PY - 1996

Y1 - 1996

N2 - Parallel simulation of fluid power systems using the transmission-line modeling method offers the benefit of increased speed of execution, but requires the system model to be partitioned on to individual processors [Burton, 1994]. In this paper we address the automatic placement of component models on processors of a distributed memory parallel machine. A genetic algorithm is used to map the different processes onto several processors. The objective is to minimise the total processing time in order to achieve real time performance. For fine grained computation problems the communication time cannot be neglected, i.e. we consider the computation and the communication time of each task. This mapping problem is a combinatorial optimization problem which can be reduced to the graph partitioning problem which is shown to be NP-complete. Combining genetic algorithms with heuristics leads to optimal or very good sub-optimal solutions. A hydraulic example circuit is partitioned in four and eight subsystems, respectively, and the simulation is implemented on a Transputer based platform. Using eight processors a speed up of 3.7 was achieved

AB - Parallel simulation of fluid power systems using the transmission-line modeling method offers the benefit of increased speed of execution, but requires the system model to be partitioned on to individual processors [Burton, 1994]. In this paper we address the automatic placement of component models on processors of a distributed memory parallel machine. A genetic algorithm is used to map the different processes onto several processors. The objective is to minimise the total processing time in order to achieve real time performance. For fine grained computation problems the communication time cannot be neglected, i.e. we consider the computation and the communication time of each task. This mapping problem is a combinatorial optimization problem which can be reduced to the graph partitioning problem which is shown to be NP-complete. Combining genetic algorithms with heuristics leads to optimal or very good sub-optimal solutions. A hydraulic example circuit is partitioned in four and eight subsystems, respectively, and the simulation is implemented on a Transputer based platform. Using eight processors a speed up of 3.7 was achieved

M3 - Article

VL - 3

SP - 83

EP - 91

JO - American Society of Mechanical Engineers, The Fluid Power and Systems Technology Division (Publication) FPST

JF - American Society of Mechanical Engineers, The Fluid Power and Systems Technology Division (Publication) FPST

ER -