[Paper Review] On the LRD of the Aggregated Traffic Flows in High-Speed Computer Networks

Ginno Millán|arXiv (Cornell University)|Mar 5, 2021

Network Traffic and Congestion Control5 references4 citations

TL;DR

This paper investigates Long-Range Dependence (LRD) in aggregated high-speed network traffic using fractal analysis, focusing on traffic flows classified under Differentiated Services. By estimating the Hurst exponent via time series analysis of real traffic traces from a university core switch, the study quantifies burstiness and LRD across traffic classes, revealing significant self-similar behavior in aggregated flows that impacts QoS provisioning.

ABSTRACT

This paper studies and analyses the behavior of the Long-Range Dependence in network traffic after classifying traffic flows in aggregated time series. Following Differentiated Services architecture principles, the generic Quality of Service applications that requirements and use the transport control protocol, a basic classification criterion of time series is established. Using the fractal theory, the resulting time series are analyzed. The Hurst exponent is estimated and used as a measure of traffic burstiness and Long-Range Dependency in each traffic class. The traffic volume per class is also measured. The study uses traffic traces collected at the core switch at the Electric Engineering Department at Universidad de Santiago de Chile in different periods of network activity.

Motivation & Objective

• To analyze Long-Range Dependence (LRD) in aggregated network traffic flows within high-speed computer networks.
• To classify traffic flows based on Differentiated Services architecture and Quality of Service (QoS) requirements.
• To quantify traffic burstiness and LRD using fractal theory and Hurst exponent estimation.
• To measure traffic volume per class and assess its impact on network behavior.

Proposed method

• Traffic traces were collected from the core switch at the Universidad de Santiago de Chile's Electrical Engineering Department.
• Traffic flows were classified into distinct classes based on QoS requirements and transport protocol usage.
• Time series of aggregated traffic per class were analyzed using fractal theory.
• The Hurst exponent was estimated to measure long-range dependence and burstiness in each traffic class.
• Statistical analysis was applied to evaluate self-similarity and correlation structure across different network activity periods.
• The study used real-world data from multiple time periods to assess temporal stability of LRD characteristics.

Experimental results

Research questions

• RQ1 How does Long-Range Dependence manifest in aggregated traffic flows across different QoS classes in high-speed networks?
• RQ2 To what extent does the Hurst exponent vary across traffic classes, indicating differing levels of burstiness and correlation?
• RQ3 What is the relationship between traffic volume per class and the degree of long-range dependence?
• RQ4 How stable is the LRD behavior across different network activity periods?
• RQ5 Can fractal analysis effectively characterize and differentiate traffic classes based on their self-similar properties?

Key findings

• The Hurst exponent estimates confirmed the presence of Long-Range Dependence in all aggregated traffic classes, indicating persistent burstiness over long time scales.
• Higher Hurst values were observed in certain traffic classes, suggesting stronger correlation structures and more pronounced self-similarity.
• Traffic volume per class showed a measurable correlation with the degree of LRD, with higher-volume classes often exhibiting more persistent patterns.
• The LRD characteristics remained relatively stable across different network activity periods, indicating consistent self-similar behavior.
• The application of fractal analysis via Hurst exponent estimation proved effective in distinguishing traffic classes based on their long-term correlation properties.