Wireless communication networks are facing a paradigm shift. From providing only voice communication, new generations of wireless networks are designed to provide different types of multimedia communications together with different types of data services and aim to seamlessly integrate in the big Internet infrastructure. Some of these applications and services have strong resource requirements in order to function properly (e.g. videoconferences), others are flexible enough to adapt to whatever is available (e.g. FTP). Also, different services (or different users), might have different importance levels, and should be treated accordingly. Providing resource assurance and differentiation is often referred to as quality of service (QoS). Moreover, due to the constrained and fluctuating bandwidth of the wireless link, and user mobility, wireless networks represent a class of distributed systems with a higher degree of unpredictability and dynamic change as compared to their wireline counterparts.

In this thesis we study how novel resource allocation algorithms can improve the behaviour (the offered QoS) of dynamic unpredictable distributed systems, such as a wireless network, during periods of overload. This work concerns both low level enforcement mechanisms and high-level policy dependent optimisation algorithms.

First, we propose and evaluate adaptive admission control algorithms for controlling the load on a processor in a radio network controller. We use feedback mechanisms inspired by automatic control techniques to prevent CPU overload, and policy-dependent deterministic algorithms to provide service differentiation.

Second, we propose and evaluate a QoS-aware bandwidth admission control and allocation algorithm for the radio link in a network cell. The acceptable quality levels for a connection are specified using bandwidth dependent utility functions, and our scheme aims to maximise system-wide utility. The novelty in our approach is that we take into account bandwidth reallocation, which arise as a consequence of the dynamic environment, and their effects on the accumulated utility of the different connections.