A measurement study on wireless networks: normal case and degraded case
Songwu Lu, Xiaoqiao Meng · 2006
The proliferation of cellular networks and wireless LANs stimulates massive efforts to understand how these systems work. In practice, due to many uncertainties, a wireless system can operate in either the normal case or degraded case. While in the normal case, the network system functions orderly and achieves expected performance, in the degraded case, which usually not occurs frequently, the network either has poor performance or ceases to function. In this thesis, by collecting data traces from real wireless networks, we analyze behavior of wireless systems (or services) in both the normal and degraded case. We start from characterizing normal-case behavior of a Short Message Service (SMS) provided by a nationwide cellular network. The studied behavior includes communication quality and traffic characteristics. The communication quality of SMS is shown not as good as it is commonly expected---both the message failure ratio and latency are high. We further present SMS traffic statistics and examine the store-and-forward aspects of the SMS. In the other popular wireless network, namely, wireless LANs, our normal-case study is focused on TCP flows. We statistically characterize both static and roaming flows in a large campus wireless network. A two-tier approach is proved to be accurate for capturing flow arrivals. We further find that the flow arrivals in spatial proximity show strong correlation. We explain the results by user behavior and application demands, and further cross-validate the results by other traces. For both SMS and TCP flows, we use examples to show that our fitted traffic models are applicable for the purpose of accurately evaluating protocol design. In the second part of this thesis, we present two illustrations of degraded-case study: SMS and base-station scheduling in cellular networks. We first identify and analyze a stressful condition for SMS, that is, a flash crowd event during New Year's Eve. We further find two factors that may potentially degrade SMS performance: bulk message delivery which can induce network congestion, and the SMS user connection structure that can pose severe threat in the future via quickly propagating viruses or spams. Solutions to avoiding these three types of degraded case will be discussed. Our study described above demonstrates that measurement is an effective approach to understanding network behavior. Nevertheless, it might not be applicable for estimating the lower bound of network performance in certain degraded case. As the last contribution of this thesis, we describe a zero-sum game theoretic approach. The approach is exemplified by the base-station scheduling issue in cellular networks, wherein the base-station needs to schedule transmissions of multiple data flows. In order to characterize the scheduler's performance when it is in a degraded case due to wireless channel errors, we assume the scheduler and the channel error play a zero-sum game. The game is then formulated into a minimax optimization problem that can be solved to give the lower bound of the degraded-case performance.