Evolutionary Approach to Maxmin-Fair Network-Resource Allocation

This paper presents an evolutionary approach to the finding of maxmin-fair solutions of the general network resource allocation problem. From the definition of maxmin-fairness, a general relation among real points is derived that can be used to replace the role of the Pareto-dominance relation in evolutionary multi-objective search algorithms. The approach has been demonstrated by modifying the SPEA2 algorithm in this manner and testing it on some problems with known fairness points. In all the studied cases, the algorithm could achieve the correct fairness points, and the search effort scaled moderately with increasing problem dimension. Since maxmin-fair points also belong to the Pareto-front, the approach is also of interest for multi-objective optimization in general.