Hypermeshes: Implementation and performance

S. Loucif*, M. Ould-Khaoua, A. Al-Ayyoub

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

15 Citations (Scopus)

Abstract

Common multicomputer networks, including the torus, mesh, and hypercube, are graph topologies where a channel interconnects exactly two nodes. Hypergraphs are generalisations of the graph model, where a channel interconnects an arbitrary number of nodes. The spanning-bus hypercube is a well-known network that belongs to the hypergraph family. Regular multidimensional hypergraphs, also known as hypermeshes, have been proposed as potential alternatives to traditional graph networks for future multicomputers due to their superior topological and performance features. This paper compares the different schemes that have been proposed in the literature for implementing the hypermesh. The results reveal that one particular version of the hypermesh, known as the distributed crossbar switch hypermesh, provides the best performance when implementation costs are taken into account.

Original languageEnglish
Pages (from-to)37-47
Number of pages11
JournalJournal of Systems Architecture
Volume48
Issue number1-3
DOIs
Publication statusPublished - Sept 2002
Externally publishedYes

Keywords

  • Graphs
  • Hypergraphs
  • Interconnection networks
  • Multicomputers
  • Parallel processing
  • Performance analysis
  • Virtual cut-through

ASJC Scopus subject areas

  • Software
  • Hardware and Architecture

Fingerprint

Dive into the research topics of 'Hypermeshes: Implementation and performance'. Together they form a unique fingerprint.

Cite this