On the Accuracy of Sampling Schemes for Wireless Network Characterization

Wireless network characterization is an important task in next generation wireless networks. In order to achieve efficient wireless network characterization, accurate sampling strategies are required. The relative performance of different sampling strategies for assessing various wireless network traffic metrics is significant due to the complexity and expense involved in the collection, storage, and analysis of all the traffic generated in the wireless medium. Since the spectrum used for most wireless networks, especially those based on IEEE 802.11 standards, is divided into several channels, the existing count-based sampling methods demand continuous capture on each channel for selecting the desired packets of interest. Continuous capturing makes the cost of monitoring infrastructure very expensive and hence count-based sampling methods are not scalable. However, the time-based sampling methods which were considered inaccurate in wired network characterization, appear to offer a cost-effective and scalable solution by reducing the cost of resources necessary to accurately characterize the wireless medium. For example, the use of time-based sampling enable us to make use of a single wireless interface for accurately sampling multiple channels. However, in order to achieve this, we need to identify the right set of parameters for time-based sampling. This paper presents a study of the performance of various time-based sampling methods in answering questions related to their use in wireless network traffic characterization. We simulate time-based sampling traces at a variety of granularities using a complete packet trace (i.e., parent population) captured in a campus wireless network environment that aggregates traffic from a large number of nodes. From our analysis using Chi-square test, we found that the timer-driven time-based sampling is more accurate than count-driven time-based sampling for both systematic and stratified sampling schemes.