| This dissertation argues that wireless LAN technology alone, which includes wireless access points for the infrastructure and network adapters for end hosts, is insufficient for providing connectivity in Public-Area Wireless Networks (PAWNs), and a number of requirements need to be met before wireless LAN connectivity can be extended beyond private networks to airports, shopping malls, hotels, and other public places. First, network administrators need to implement mechanisms for protecting the network against unknown, and potentially malicious users, and protecting user communication traffic. Second, as PAWNs become more widely deployed, it is crucial to consider techniques that effectively handle the dynamically varying user load without compromising network performance. Third, as PAWN traffic becomes more multimedia-centric, the need to address quality of service issues is increasingly important.; In this dissertation, we address these challenges in implementing and deploying PAWNs by extending previous research in three significant ways: (i) we propose a centralized PAWN architecture that supports mechanisms for wireless-hop security, capacity planning, bandwidth management and monitoring, and location-aware services, (ii) we incorporate knowledge of user location in a novel way in the network layer to address issues in PAWN deployment, such as capacity planning, and performance optimizations, such as load balancing across access points, and wireless-hop bandwidth allocation, and (iii) by analyzing real wireless LAN workloads, we propose and evaluate two new approaches for inter-AP load balancing and wireless-hop bandwidth allocation in PAWNs.; First, we use system design, implementation, measurement, trace-driven simulations, and analytic techniques to explore the performance benefits of incorporating these mechanisms in PAWNs. Second, we characterize workloads in a wireless network setting, and then explicitly incorporate our results in addressing issues of capacity planning, load balancing, and wireless-hop bandwidth allocation. We then present two novel algorithms that address load balancing and bandwidth allocation using a combined solution—workload and location-aware adaptation. We use trace-driven simulation to demonstrate the performance advantages of incorporating user workloads and location in balancing load and in allocating wireless-hop bandwidth. And finally, we develop an API for real-time RF network sensing, and use it to implement our load balancing algorithm in a prototype system. |
