Traffic Characterization And Prediction In IEEE 802.11 WLANs

The study of network traffic characteristics is not only important for network performance evaluation but also imperative for the proper network design, management, and control. In fact, for IEEE 802.11 based Wireless Local Area Networks (WLANs), the knowledge of traffic characteristics is essential in improving the overall network Quality of Service (QoS), the performance of various IEEE 802.11 based protocols, and the design of a variety of network devices. For example, this knowledge can help us devise some advanced MAC schemes and buffer scheduling algorithm, which would achieve greater efficacy in WLAN environments. Moreover, the efficiency of congestion control, admission control, and dynamic bandwidth allocation in WLAN can be further enhanced by the accurate prediction of traffic, allowing for better utilization of wireless resources. Therefore, with the flourishing development of wireless networks, the research on traffic characterization and prediction has great significance in IEEE 802.11 based WLANs.In this dissertation, we first investigated the statistical characteristics of packet inter-arrival times and aggregated wireless traffic in WLAN. We found that the inter-arrival times are indeed asymptotically self-similar to long-range dependent (LRD) traffic. In this case, the traditional Poisson distribution is not appropriate in describing the inter-arrival times in WLAN. Additionally, the modeling methods used in wired networks cannot be applied to WLAN performance analyses directly. The Augmented-Dickey-Fuller (ADF) tests indicated that aggregated wireless traffic series were non-stationary and all of them were integrated of order one I(1). We also studied the second-order scaling properties of the aggregated traffic and found that upstream and downstream aggregated traffics were weakly self-similar at small timescales (sub-second timescales) but self-similar at large timescales. Compare to Poisson process, the wavelet energy plot of aggregated traffic shows that the aggregated traffic has higher energy (more bursty) than Poisson process for all scales.Secondly, we applied Chaos Theory to characterize aggregated wireless traffic. Wireless traffic was analyzed qualitatively and the power spectrum showed differences among wireless traffic, random time series and periodic time series. The principal component showed that all wireless traffic streams had nonlinear characteristics; and chaos had been found in all aggregated wireless traffic via power spectrum and principal content analysis. In order to confirm the chaotic characteristic of wireless traffic, the correlation dimension and the Lyapunov exponents method are used in the quantitative analysis of wireless traffic as tests for chaos. The calculation results show that all wireless traffic has non integer correlation dimension. Their maximum Lyapunov exponents are positive numbers. All of the results above provided prove that the aggregated wireless traffic is chaotic.