| Efficient and reliable bandwidth allocation is one of the most important open issues in the management of networks that aim to offer a guaranteed Quality of Service. The bandwidth allocation problem becomes more difficult in multi-service networks, where a large variety of different applications, each one with different requirements in terms of bandwidth, duration or delay, information loss use the network infrastructure simultaneously. Most of the previous work has analyzed bandwidth allocation policies under the context of resolving conflicts due to dynamics of user requests without taking network availability, user mobility, or the delivery (i.e., physical environment) conditions into account. Since static bandwidth allocation policies lack adaptive mechanisms to combat these dynamics in the network and improve bandwidth utilization, we believe that a more flexible service model which allows variable QoS is needed. Adaptive resource management coupled with dynamic load balancing aims at decreasing the possibility of congestion and maintaining high resource utilization, under transient traffic variations and node/link failure. We have formulated preemption algorithms and criteria for optimization by preemption algorithms, studied existing algorithms and investigated sub-optimal preemption algorithms with random selection of connections to be rerouted. We have also performed numerical and simulation comparisons of re-routing algorithms by analyzing their performance on a single link, dynamic setting and in a full network environment with a heterogeneous traffic mix. In order to account for the users application type, QoS requirements and quantify users' value we have introduced a utility-based QoS model. We have investigated network utilization, QoS observed by the customers, and revenue generation perspectives for different utility-quantified bandwidth allocation schemes. We have presented approximate analytical tools to obtain blocking probabilities in a multi rate multi class system, where users of the same class can have different resource requirements. We have evaluated the blocking probabilities for a single link case and validated our approach through the simulation of such a system. Also we have expanded our single link model to calculate blocking probabilities for a multihop path, when the offered traffic of each source destination pair along the path is known. |
