The theme of this thesis is the design and the analysis of control mechanisms that allow the bandwidth sharing properties of the best effort Internet to be changed in a well-defined and predictable manner. We focus on end-to-end mechanisms, where the end-points are user agent processes running on the hosts. In a conscious attempt to minimise the dependencies on routers and opt for simplicity and scalability the routers maintain the traditional First Come First Serve (FCFS) service discipline. Meanwhile, the end-points are allowed to select from a predefined range of classes with certain control parameters, which carry specific weights and are known to have proportional bandwidth sharing capabilities. Using a deterministic, discrete-time model, we first study the feasibility of such a distributed control scheme, derive the parameter relationships necessary for optimal control and prove system stability. Then we provide an analysis of system efficiency and weighted fairness for a range of control parameters and class populations. Furthermore, packet level simulations and real network experiments with appropriately modified congestion control in the Transmission Control Protocol (TCP), showed that strict weighted fair bandwidth sharing is possible in practice between connections competing simultaneously for bandwidth in the same end-to-end network path but for a limited range of proportional weights and only under modest congestion conditions.