Autonomic wireless networking

Large-scale deployment of IEEE 802.11 wireless LANs (WLANs) remains a significant challenge. Many access points (APs) must be deployed and interconnected without a-priori knowledge of the demand. We consider that the deployment should be iterative, as follows. At first, access points are deployed to achieve partial coverage. Then, usage statistics are collected while the network operates. Overloaded and under-utilized APs would be identified, giving the opportunity to relocate, add or remove APs. In this thesis, we propose extensions to the WLAN architecture that would make our vision of iterative deployment feasible. One line of work focuses on self-configuration, which deals with building a WLAN from APs deployed without planning, and coping with mismatches between offered load and available capacity. Self-configuration is considered at three levels. At the network level, we propose a new distribution system that forms a WLAN from a set of APs connected to different IP networks and supports AP auto-configuration, link-layer mobility, and sharing infrastructure between operators. At the inter-cell level, we design a load-balancing scheme for overlapping APs that increases the network throughput and reduces the cell delay by evenly distributing the load. We also suggest how to reduce the handoff time by early detection and fast active scanning. At the intra-cell level, we present a distributed admission control that protects cells against congestion by blocking stations whose MAC service time would be above a set threshold. Another line of work deals with self-deployment and investigates how the network can assist in improving its continuous deployment by identifying the reasons for low cell throughput. One reason may be poor radio conditions. A new performance figure, the Multi-Rate Performance Index, is introduced to measure the efficiency of radio channel usage. Our measurements show that it identifies cells affected by bad radio conditions. An additional reason may be limited performance of some AP models. We present a method to measure the upper bound of an AP’s throughput and its dependence on offered load and orientation. Another reason for low throughput may be excessive distance between users and APs. Accurate positioning of users in a WLAN would permit optimizing the location and number of APs. We analyze the limitations of the two most popular range estimation techniques when used in WLANs: received signal strength and time of arrival. We find that the latter could perform better but the technique is not feasible due to the low resolution of the frame timestamps in the WLAN cards. The combination of self-configuration and self-deployment enables the autonomic operation of WLANs