Publication No 36365

Author(s)

Hu, G.*

Title

Impact of Access Bandwidth on Aggregated Traffic Behavior and Queueing Performance

Topics

Broadband Networks; Internet

Methods

Measurement and Trials; Performance Evaluation

Keywords

TRAFFIC ANALYSIS; TRAFFIC MODEL; LONG RANGE DEPENDENCE; AGGREGATION; SIMULATION; QUEUEING SYSTEM; PERFORMANCE EVALUATION

Abstract

Recent traffic measurement of IP backbone networks discovered that aggregated IP traffic can be either uncorrelated or strongly correlated at small time scales, although at large time scales it exhibits long rang dependence (LRD). Based on the infinite source Poisson traffic model, we show in this paper that the lack of correlation is an intrinsic small time scale property of LRD traffic due to the multiplexing of a large number of independent traffic flows. Particularly, the access bandwidth of users is a critical factor determining the boundary between the two ranges of time scales in which the uncorrelation and correlation property dominate the traffic behaviour respectively. A higher access bandwidth makes this boundary located at a smaller time scale. We refer to this time scale as boundary time scale and argue that the existence of a very small boundary time scale can explain the reported strong correlation at small time scales. Furthermore, the traffic behavior at different time scales leads to different queueing behaviors with respect to different queueing lengths. It is shown a large boundary time scale results in a large degree of buffer efficiency and thus brings substantial performance improvement in spite of the existence of LRD at large time scales.

Year

2004

Reference entry

Hu, G.
Impact of Access Bandwidth on Aggregated Traffic Behavior and Queueing Performance
Proceedings of the 12th GI/ITG Conference on Measuring, Modelling and Evaluation of Computer and Communication Systems/3rd Polish-German Teletraffic Symposium, Dresden, September 2004

BibTex file

Download  [BIBTEX]

Full Text

Download  [PDF]