Speedup and synchronization overhead analysis of Gauss-Seidel type algorithms on a Sequent balance machine

Garng Morton Huang, W. Ongsakul

Research output: Contribution to journalArticle

7 Citations (Scopus)

Abstract

The parallelization and implementation of Gauss-Seidel (G-S) algorithms for power flow analysis have been investigated previously. Numerous runs to demonstrate the speedup have been illustrated on a Sequent Balance shared-memory multi-instruction, multidata access (SM MIMD) machine. The authors extend the idea and investigate the effects of acceleration factors. It is shown on systems ranging from teens to thousands that when the acceleration factors are used, the implementation using colour-by-colour synchronization is more reliable and has better convergence rate, even though it takes longer time to synchronize. The authors also analyze the dependence of synchronization overhead in terms of system sizes, network connection and number of processors. Comparisons between G-S and fast decoupled load-flow algorithms are also made. The implications on nCUBE implementations are also discussed. It is also shown that the idea of parallel G-S algorithm can be easily extended to solve the transient stability problem which involves a set of algebraic differential equations.

Original languageEnglish
Pages (from-to)437-444
Number of pages8
JournalIEE Proceedings: Generation, Transmission and Distribution
Volume141
Issue number5
DOIs
Publication statusPublished - 1 Sep 1994
Externally publishedYes

Fingerprint

Synchronization
Color
Differential equations
Data storage equipment

ASJC Scopus subject areas

  • Electrical and Electronic Engineering

Cite this

@article{354be72991d34037a90a8a51de399d14,
title = "Speedup and synchronization overhead analysis of Gauss-Seidel type algorithms on a Sequent balance machine",
abstract = "The parallelization and implementation of Gauss-Seidel (G-S) algorithms for power flow analysis have been investigated previously. Numerous runs to demonstrate the speedup have been illustrated on a Sequent Balance shared-memory multi-instruction, multidata access (SM MIMD) machine. The authors extend the idea and investigate the effects of acceleration factors. It is shown on systems ranging from teens to thousands that when the acceleration factors are used, the implementation using colour-by-colour synchronization is more reliable and has better convergence rate, even though it takes longer time to synchronize. The authors also analyze the dependence of synchronization overhead in terms of system sizes, network connection and number of processors. Comparisons between G-S and fast decoupled load-flow algorithms are also made. The implications on nCUBE implementations are also discussed. It is also shown that the idea of parallel G-S algorithm can be easily extended to solve the transient stability problem which involves a set of algebraic differential equations.",
author = "Huang, {Garng Morton} and W. Ongsakul",
year = "1994",
month = "9",
day = "1",
doi = "10.1049/ip-gtd:19941210",
language = "English",
volume = "141",
pages = "437--444",
journal = "IEE Proceedings: Generation, Transmission and Distribution",
issn = "1350-2360",
publisher = "Institute of Electrical Engineers",
number = "5",

}

TY - JOUR

T1 - Speedup and synchronization overhead analysis of Gauss-Seidel type algorithms on a Sequent balance machine

AU - Huang, Garng Morton

AU - Ongsakul, W.

PY - 1994/9/1

Y1 - 1994/9/1

N2 - The parallelization and implementation of Gauss-Seidel (G-S) algorithms for power flow analysis have been investigated previously. Numerous runs to demonstrate the speedup have been illustrated on a Sequent Balance shared-memory multi-instruction, multidata access (SM MIMD) machine. The authors extend the idea and investigate the effects of acceleration factors. It is shown on systems ranging from teens to thousands that when the acceleration factors are used, the implementation using colour-by-colour synchronization is more reliable and has better convergence rate, even though it takes longer time to synchronize. The authors also analyze the dependence of synchronization overhead in terms of system sizes, network connection and number of processors. Comparisons between G-S and fast decoupled load-flow algorithms are also made. The implications on nCUBE implementations are also discussed. It is also shown that the idea of parallel G-S algorithm can be easily extended to solve the transient stability problem which involves a set of algebraic differential equations.

AB - The parallelization and implementation of Gauss-Seidel (G-S) algorithms for power flow analysis have been investigated previously. Numerous runs to demonstrate the speedup have been illustrated on a Sequent Balance shared-memory multi-instruction, multidata access (SM MIMD) machine. The authors extend the idea and investigate the effects of acceleration factors. It is shown on systems ranging from teens to thousands that when the acceleration factors are used, the implementation using colour-by-colour synchronization is more reliable and has better convergence rate, even though it takes longer time to synchronize. The authors also analyze the dependence of synchronization overhead in terms of system sizes, network connection and number of processors. Comparisons between G-S and fast decoupled load-flow algorithms are also made. The implications on nCUBE implementations are also discussed. It is also shown that the idea of parallel G-S algorithm can be easily extended to solve the transient stability problem which involves a set of algebraic differential equations.

UR - http://www.scopus.com/inward/record.url?scp=0028510745&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0028510745&partnerID=8YFLogxK

U2 - 10.1049/ip-gtd:19941210

DO - 10.1049/ip-gtd:19941210

M3 - Article

AN - SCOPUS:0028510745

VL - 141

SP - 437

EP - 444

JO - IEE Proceedings: Generation, Transmission and Distribution

JF - IEE Proceedings: Generation, Transmission and Distribution

SN - 1350-2360

IS - 5

ER -