SWelf-Similarity in Wireless Traffic

Motivation and Goal

In recent years, researchers have convincingly shown that traffic in both wired LANs and wired WANs exhibit the feature of self-similarity (a different yet equivalent mathematical manifestation of long-range dependence). A natural question to ask is, will wireless mobile traffic, which typically run the same set of applications but on mobile hosts over wireless channel, also exhibit such a feature?
The project is intended to address this problem. Specifically, we try to answer the following questions

  • What's the traffic characteristic in wireless LAN? Does it also show self-similarity?
  • Do mobility and channel error have impact on the traffic characteristic?
  • Does the wireless traffic characteristic affect on network design?

Our Approaches and Results

  • We use 3-months real traffic traces collected from a WaveLAN. These traces show that the aggregate wireless traffic exhibits self-similarity up to a certain time scale. Over this time scale, it is second-order non-stationary, thus it becomes not appropriate to still use self-similarity to model it. The traffic also shows diurnal periodicity.
  • Mobility will change the number of active flows in the cell. However, when the traffic is high, mobility will not change the self-similarity. We also show the widely used Hong/Rappaport mobility model does not affect self-similarity.
  • We use real  variant-bit-rate (VBR) video trace to drive ns-2 as the background traffic, then we impose real channel error patterns on the video traffic. As we know, VBR video traffic is self-similar. The simulation shows that channel errors do not affect self-similarity.
  • Self-similarity means scale-invariant burstiness. It has been demonstrated as harmful for networking designs such as queueing performance, buffer reservation. We implement two measurement-based admission control methods in simulators, i.e., Time-window based and Exponential averaging. The results show that self-similarity will bring the wireless link utilization to less than 66%, and it will break down Time-window based admission control.

Project Members

  • Student: Xiaoqiao Meng, Hao Yang
  • Faculty: Songwu Lu

Publications

 
WiNG