Result Details

Design of Low-Cost Communication Algorithms for Irregular Networks

STAROBA, J.; DVOŘÁK, V. Design of Low-Cost Communication Algorithms for Irregular Networks. Proceedings of the 3rd International Conference on Networking ICN '04. Colmar: University of Haute Alsace, 2004. p. 980-985. ISBN: 0-86341-325-0.
Type
conference paper
Language
English
Authors
Staroba Jiří, Ing., Ph.D.
Dvořák Václav, prof. Ing., DrSc., DCSY (FIT)
Abstract

Communication between non-adjacent nodes in irregular interconnection networks mostly relies on routing tables. Efficient implementation of group communication algorithms such as total exchange also needs similar tables. Since such tables cannot be derived mathematically, we have used a parallel genetic algorithm GAroute to design a sub-optimal routing schedule for a sequence of communication steps (neighbor-to-neighbor transfers) during group communications. Genetic evolution does not necessarily lead to the minimum number of steps, but provides satisfactory low-cost communication algorithms. Accuracy of obtained results has been tested with a regular hypercube topology with known complexity of communication.

Keywords

communications, genetic algorithms, communication architectures, parallel processing

Published
2004
Pages
980–985
Proceedings
Proceedings of the 3rd International Conference on Networking ICN '04
Conference
3rd International Conference on Networking
ISBN
0-86341-325-0
Publisher
University of Haute Alsace
Place
Colmar
BibTeX
@inproceedings{BUT16907,
  author="Jiří {Staroba} and Václav {Dvořák}",
  title="Design of Low-Cost Communication Algorithms for Irregular Networks",
  booktitle="Proceedings of the 3rd International Conference on Networking ICN '04",
  year="2004",
  pages="980--985",
  publisher="University of Haute Alsace",
  address="Colmar",
  isbn="0-86341-325-0"
}
Projects
Parallel system performance prediction and tuning, GACR, Standardní projekty, GA102/02/0503, start: 2002-01-01, end: 2004-12-31, completed
Research groups
Departments
Back to top