Architecting adaptive resource-aware protocols for large wireless networks

Capacity in a wireless network is fundamentally limited by the amount of spectrum available. The surge in mobile hand-held devices such as smartphones, tablets and laptops as well as the popularity of high bandwidth applications such as photo sharing and video streaming are stretching the limits of the existing network deployments. Owing to the flexibility and low cost, WiFi networks are becoming the primary means for these devices to gain access to the Internet. The problems of limited spectrum and interference caused by dense deployment and non-802.11 compliant devices (such as baby monitors and microwaves) are posing a serious problem to the link capacities achievable in practice. As a consequence, it is common for users to experience congestion and performance degradation on wireless links.;We argue that one of the primary causes of congestion and network degradation is that the wireless networks are not agile enough to adapt to the network and usage conditions. We believe that in order to support the increasing number of wireless devices and applications with high capacity demands, it is critical to design protocols that can measure the network conditions, estimate requirements and adapt accordingly. The objective of this dissertation is to evolve large-scale wireless networks such that network protocols are able to measure the available resources, analyze the network usage and operational environment and adapt dynamically based on network usage and resource availability.;At the foundation of adaptive systems are sound measurement and monitoring systems that are accurate yet lightweight. We argue that active and passive measurement studies of operational networks not only enable design of adaptive protocols but also help identify fundamental deficiencies in existing protocols. To this end, we build measurement infrastructure and analyze the operation of two diverse networks: large-scale WiFi deployment in a conference and residential WiFi networks. Our work exposes the deficiency in 802.11 protocol that results in sub-optimal performance when the network usage increases.;Due to the inherent broadcast nature of the wireless medium, the links tend to be loss prone and can experience high latency. In order to improve performance, several knobs are available to a protocol developer, such as modulation, transmit power and center frequency. We explore two fundamental knobs---spectrum awareness and capacity awareness. Most wireless communications today involve the use of channels with preset widths. In this work, we propose that devices should adapt another knob that is relatively unexplored, namely the channel width, based on demand and channel conditions in order to improve communication. Secondly, we make a case for aggressive wireless link utilization in order to aggregate the low capacity Internet backhauls of broadband networks.;Finally, in order to deploy and maintain a large-scale network, it is important to have a lightweight yet effective management system. Prior monitoring and management systems tend to one of the extremes of collecting too little data or too much. We present a monitoring system that adapts the granularity of data collection based on the network health. In a well functioning network, a small amount of metrics collection is sufficient, but as and when problems occur, the data collection scales to aid problem detection. Such adaptive monitoring systems not only reduce the management overhead but also aid fault diagnosis and real time network visualization.;The key problems identified in this dissertation and the solutions proposed are not only applicable to retrofitting solutions in WiFi networks, but also in designing networks in other parts of the spectrum where the channel widths can be varied. We believe that the measurement studies and the solutions proposed in this dissertation will help realize the goal of operating networks of large scale.