Abstract |
Wireless local area networks are increasingly being deployed in a wide variety of areas to meet the growing demand for wireless access. When compared to their wired counterparts, wireless networks experience larger delays, lower throughput and more frequent packets losses and retransmissions. Thus, developing mechanisms, such as, load balancing, capacity planning, and admission control to improve their performance will become more and more important. In this context, it is critical to understand and analyze the performance and workload characteristics of wireless networks in order to develop wireless networks that are more scalable, robust, easier to manage and able to utilize their scarce resources more efficiently. Moreover, empirical studies can be used to guide modeling efforts for wireless demand and access patterns and provide realistic models to performance analysis studies for wireless network protocols and services.
In this work, we study a large wireless infrastructure and explore the characteristics of traffic load at APs in order to derive simple models for it. We carry measurements based on traces acquired from the wireless infrastructure of the University of North Carolina (UNC). In the first part of the study, we perform a statistical analysis on the wireless traffic load of APs to derive models that can be used in traffic load forecasting. Based on these models, we design and evaluate several simple forecasting algorithms. The traffic characterization is extended by classifying flows into application types using a graph-based method in order to avoid the inherent limitations of a port-based classification. We perform an application-based characterization of traffic at different spatial scales and compare this methodology with others in literature. Our group has proposed models for the wireless traffic demand using real-life traces collected from the wireless infrastructure at UNC. The
final part of this thesis discusses a synthetic trace generator that produces
synthetic traffic based on these models. By replaying such traces,
researchers could analyze the performance of various wireless networking
protocols under realistic traffic conditions.
|