Affordable Access

Access to the full text

Identifying elephant flows using dynamic thresholds in programmable IXP networks

Authors
  • Silva, Marcus Vinicius Brito da1, 2
  • Marques, Jonatas Adilson1
  • Gaspary, Luciano Paschoal1
  • Granville, Lisandro Zambenedetti1
  • 1 Institute of Informatics, Federal University of Rio Grande do Sul, Porto Alegre, Brazil , Porto Alegre (Brazil)
  • 2 Federal Institute of Pará, IFPA, Cametá, Brazil , Cametá (Brazil)
Type
Published Article
Journal
Journal of Internet Services and Applications
Publisher
Springer London
Publication Date
Dec 10, 2020
Volume
11
Issue
1
Identifiers
DOI: 10.1186/s13174-020-00131-6
Source
Springer Nature
Keywords
License
Green

Abstract

Internet eXchange Points (IXPs) are Internet infrastructures composed of high-performance networks that allow multiple autonomous systems to exchange traffic. Given the challenges of managing the flows that cross an IXP, identifying elephant flows may help improve the quality of services provided to its participants. In this context, we leverage the new flexibility and resources of programmable data planes to identify elephant flows in IXP networks adaptively via the dynamic adjustment of thresholds. Our mechanism uses the information reported by the data plane to monitor network utilization in the control plane, calculating new thresholds based on previous flow sizes and durations percentiles and configuring them back into switches to support the local classification of flows. Thus, the thresholds are updated to make the identification process better aligned with the network behavior. The experimental results show that it is possible to identify and react to elephant flows quickly, less than 0.4ms, and efficiently, with only 98.4KB of data inserted into the network by the mechanism. In addition, the threshold updating mechanism achieved accuracy of up to 90% in our evaluation scenarios.

Report this publication

Statistics

Seen <100 times