| dc.description.abstract | The 802.11 standard has become the dominant protocol for Wireless Local Area Networks (WLANs). As an indication of its current and growing popularity, it is estimated that over 20 billion WiFi chipsets will be shipped between 2016 and 2021. In a span of less than 20 years, the speed of these networks has increased from 11 Mbps to several Gbps. The ever-increasing demand for more bandwidth required by applications such as large downloads, 4K video streaming, and virtual reality applications, along with the problems caused by interfering WiFi and non-WiFi devices operating on a shared spectrum has made the evaluation, understanding, and optimization of the performance of 802.11 networks an important research topic.
In 802.11 networks, highly variable channel conditions make conducting valid, repeatable, and realistic experiments extremely challenging. Highly variable channel conditions, although representative of what devices actually experience, are often avoided in order to conduct repeatable experiments. In this thesis, we study existing methodologies for the empirical evaluation of 802.11 networks. We show that commonly used methodologies, such as running experiments multiple times and reporting the average along with the confidence interval, can produce misleading results in some environments.
We propose and evaluate a new empirical evaluation methodology that expands the environments in which repeatable evaluations can be conducted for the purpose of comparing competing alternatives. Even with our new methodology, in environments with highly variable channel conditions, distinguishing statistically significant differences can be very difficult because variations in channel conditions lead to large confidence intervals. Moreover, running many experiments is usually very time consuming. Therefore, we propose and evaluate a trace-based approach that combines the realism of experiments with the repeatability of simulators. A key to our approach is that we capture data related to properties of the channel that impact throughput. These traces can be collected under conditions representative of those in which devices are likely to be used and then used to evaluate different algorithms or systems, resulting in fair comparisons because the alternatives are exposed to identical channel conditions.
Finally, we characterize the relationships between the numerous transmission rates in 802.11n networks with the purpose of reducing the complexities caused by the large number of transmission rates when finding the optimal combination of physical-layer features. We find that there are strong relationships between most of the transmission rates over extended periods of time even in environments that involve mobility and experience interference. This work demonstrates that there are significant opportunities for utilizing relationships between rate configurations in designing algorithms that must choose the best combination of physical-layer features to use from a very large space of possibilities. | en |
