On light and heavy traffic approximations of balanced fairness

Thomas Bonald, Aleksi Penttinen, Jorma Virtamo

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Flow level analysis of communication networks with multiple shared resources is generally difficult. A recently introduced sharing scheme called balanced fairness has brought, these systems within the realm of tractability. While straightforward in principle, the numerical evaluation of practically interesting performance metrics like per-flow throughput, is feasible for limited state spaces only, besides some specific networks where the results are explicit, In the present paper, we study the behaviour of balanced fairness in light and heavy traffic regimes and show how the corresponding performance results can be used to approximate the flow throughput over the whole load range. The results apply to any network, with a state space of arbitrary dimension, A few examples are explicitly worked out, to illustrate the concepts.

Original languageEnglish
Title of host publicationSIGMETRICS 2006/Performance 2006 - Joint International Conference on Measurement and Modeling of Computer Systems, Proceedings
Pages109-119
Number of pages11
Edition1
DOIs
Publication statusPublished - 1 Jun 2006
Externally publishedYes
EventSIGMETRICS 2006/Performance 2006 - Joint International Conference on Measurement and Modeling of Computer Systems - Saint Malo, France
Duration: 26 Jun 200630 Jun 2006

Publication series

NamePerformance Evaluation Review
Number1
Volume34
ISSN (Print)0163-5999
ISSN (Electronic)0163-5999

Conference

ConferenceSIGMETRICS 2006/Performance 2006 - Joint International Conference on Measurement and Modeling of Computer Systems
Country/TerritoryFrance
CitySaint Malo
Period26/06/0630/06/06

Keywords

  • Balanced fairness
  • Elastic traffic
  • Flow level analysis
  • Throughput approximation

Fingerprint

Dive into the research topics of 'On light and heavy traffic approximations of balanced fairness'. Together they form a unique fingerprint.

Cite this