Performance modeling of n-dimensional mesh networks

Pedram Rajabzadeh, Hamid Sarbazi-Azad*, Hamid Reza Zarandi, Ebrahim Khodaie, Hashem Hashemi-Najafabadi, Mohamed Ould-Khaoua

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

3 Citations (Scopus)

Abstract

Mesh-based interconnection networks are the most popular inter-processor communication infrastructures used in current parallel supercomputers. Although many analytical models of n-D torus interconnection networks have been reported in the literature over the last decade, few analytical models have been proposed for the 2-D mesh case (and not for the general n-D mesh network) using inaccurate approximations as they have not fully incorporated the asymmetry effects of the mesh topology, in order to reduce the model complexity. There has not been reported, to the best of our knowledge, a performance model that can deal with the n-D mesh network. To fill this gap, in this paper, we propose the first analytical performance model of the n-D mesh using adaptive wormhole routing. To this end, we calculate the exact traffic rates over different network channels and determine the average message latency by averaging over the message latency values corresponding to all possible sourcedestination pairs of nodes in the network. Simulation results show that the proposed model can predict the message latency fairly accurately under various working conditions.

Original languageEnglish
Pages (from-to)1304-1323
Number of pages20
JournalPerformance Evaluation
Volume67
Issue number12
DOIs
Publication statusPublished - Dec 2010

Keywords

  • Adaptive wormhole routing
  • Analytical modeling
  • Interconnection networks
  • Multicomputers
  • Performance evaluation
  • n-D meshes

ASJC Scopus subject areas

  • Software
  • Modelling and Simulation
  • Hardware and Architecture
  • Computer Networks and Communications

Fingerprint

Dive into the research topics of 'Performance modeling of n-dimensional mesh networks'. Together they form a unique fingerprint.

Cite this