Real-Time Estimation of Self-Similarity in Wireless Traffic

Abstract

Self-Similarity (S-S) is a valuable property of various networks traffic being abundantly studied for both analysis and applications. It is promising to test the S-S property for recent network traffic and potential discoveries would contribute to more useful applications. Besides, although off-line analysis of the S-S property has predominated since 1990s, there are recent studies showing that real-time estimation of the S-S property can be applied to some tailor-made applications. For example, it may help detect some network attacks because the S-S property parameter reacts quickly to those attacks. We first demonstrate the existence of the S-S property of a recent collection of 802.11 wireless network data traces, and then present a faster implementation of an existing linear-modeling method Variance-Time Plot (VTP) for those data traces, as well as a non-linear modeling of VTP plot being examined and shown to represent the S-S property more appropriate and accurate for testing data traces. Results showed that our proposed method achieved the processing time much shorter than original VTP method, while the accuracy remained in an accepted range. This performance would enable several real-time applications which require swift response to the change of the S-S property. For the non-linear modeling of VTP plot, the accuracy was shown to be as high as the proposed modified VTP method. Meanwhile, the results showed that the new parameter set of proposed method would enhance the understanding and performance of original VTP method.