Self-Similar Ethernet LAN Traffic - PowerPoint PPT Presentation

About This Presentation
Title:

Self-Similar Ethernet LAN Traffic

Description:

The original paper on network traffic self-similarity appeared at the 1993 ACM ... variance-time plot, periodograms, Whittle's estimator, maximum likelihood ... – PowerPoint PPT presentation

Number of Views:107
Avg rating:3.0/5.0
Slides: 10
Provided by: careywil
Category:

less

Transcript and Presenter's Notes

Title: Self-Similar Ethernet LAN Traffic


1
Self-Similar Ethernet LAN Traffic
  • Carey Williamson

University of Calgary
2
Introduction
  • The original paper on network traffic
    self-similarity appeared at the 1993 ACM SIGCOMM
    Conference on Communications Architectures and
    Protocols
  • Authors Will Leland, Murad Taqqu, Walter
    Willinger, and Daniel Wilson (Leland et al. 1993)

3
Introduction (Contd)
  • Extended version of the paper appeared in the
    IEEE/ACM Transactions on Networking, Vol. 2, No.
    1, February 1994
  • One of the landmark papers of the 90s
  • Highly regarded, influential, one of the most
    cited papers in the last 3 years

4
Main Contributions
  • Identified presence of self-similarity property
    in aggregate Ethernet traffic
  • Defined methodology for testing for the presence
    of self-similarity
  • variance-time plot
  • R/S statistic
  • periodogram (power spectrum)
  • Proposed explanations/models for SS

5
Self-Similarity A Hot Topic
  • Several papers since then have identified network
    traffic self-similarity in OTHER types of traffic
    (video, Internet, Web)
  • Several models for self-similar traffic have been
    proposed in the literature
  • Several studies of the performance implications
    of self-similar traffic and long-range dependence

6
Measurement Study
  • Detailed measurement study of very lengthy
    Ethernet packet traces, with high resolution
    timer, and lots of storage space
  • One of the traces presented in their paper is a
    27.5 hour trace
  • Over 20 million packets

7
Data Analysis
  • Detailed statistical analysis
  • aggregation, autocorrelation, R/S analysis,
    variance-time plot, periodograms, Whittles
    estimator, maximum likelihood ...
  • Very rigourous confidence intervals,
    sophisticated statistical tests, sound
    methodology, ...
  • A wonderful paper to read (over and over)

8
Main Results
  • Aggregate Ethernet LAN traffic is
    self-similar
  • Burstiness across many time scales
  • Hurst parameter 0.7 lt H lt 0.9
  • H is larger when network utilization is
    higher (e.g., 0.9 when U 15)
  • Self-similarity present on all LANs tested

9
Conclusions
  • Self-similarity is present in aggregate Ethernet
    LAN traffic
  • Traffic does not aggregate well at all
  • Law of large numbers may not hold!
  • Poisson models (or Markovian models of any sort)
    do not capture reality at all
  • Important to consider self-similar traffic
Write a Comment
User Comments (0)
About PowerShow.com