Fairness mechanisms for single-path capacity allocation based on mixed-integer (Non-)linear programming

An efficient and fair network resource allocation is crucial for both operators as well as users. This paper analyses the performance and fairness properties of transport-layer based resource assignment methods for individual flows on a bottleneck link. To find an optimum trade-off between maximization of the overall bandwidth and fairness, the assignment problem is solved using (non-)linear programming expressed by the objective function and various constraints. For bottleneck fairness, a factor alpha is introduced where non-zero values penalize flows which occupy multiple bottlenecks, resulting in an increase of the overall throughput while sometimes reducing the fairness, which reflects the behaviour of TCP in multi-bottleneck scenarios. For two example scenarios, the results of the introduced objective functions and fairness methods are discussed and their properties compared quantitatively by means of well-known performance metrics. It is shown that, even though the investigated methods yield different bandwidth allocations for the individual flows, the performance metric results are similar for both fairness as well as throughput, so for complex scenarios which require a large amount of calculations, a method with a low computational effort, i. e. including only linear components can be selected.