|dc.description.abstract||Wireless networks are growing at a phenomenal rate. This growth is causing an overcrowding of the unlicensed RF spectrum, leading to increased interference between co-located devices. Existing decentralized medium access control (MAC) protocols (e.g.
IEEE 802.11a/b/g standards) are poorly designed to handle interference in such dense
wireless environments. This is resulting in networks with poor and unpredictable performance, especially for delay-sensitive applications such as voice and video.
This dissertation presents a practical conflict-graph (CG) based approach to designing self-organizing enterprise wireless networks (or WLANs) where interference is centrally managed by the network infrastructure. The key idea is to use potential interference information (available in the CG) as an input to algorithms that optimize the parameters
of the WLAN.We demonstrate this idea in three ways. First, we design a self-organizing
enterprise WLAN and show how the system enhances performance over non-CG based
schemes, in a high fidelity network simulator. Second, we build a practical system for conflict graph measurement that can precisely measure interference (for a given network configuration) in dense wireless environments. Finally, we demonstrate the practical benefits
of the conflict graph system by using it in an optimization framework that manages
associations and traffic for mobile VoIP clients in the enterprise.
There are a number of contributions of this dissertation. First, we show the practical
application of conflict graphs for infrastructure-based interference management in dense wireless networks. A prototype design exhibits throughput gains of up to 50% over traditional approaches. Second, we develop novel schemes for designing a conflict graph measurement system for enterprise WLANs that can detect interference at microsecond-level
timescales and with little network overhead. This allows us to compute the conflict
graph up to 400 times faster as compared to the current best practice proposed in the
literature. The system does not require any modifications to clients or any specialized
hardware for its operation. Although the system is designed for enterprise WLANs, the
proposed techniques and corresponding results are applicable to other wireless systems as well (e.g. wireless mesh networks). Third, our work opens up the space for designing novel fine-grained interference-aware protocols/algorithms that exploit the ability to compute the conflict graph at small timescales. We demonstrate an instance of such a system with the design and implementation of an architecture that dynamically manages client associations and traffic in an enterprise WLAN. We show how mobile clients sustain uninterrupted and consistent VoIP call quality in the presence of background interference for the duration of their VoIP sessions.||en