Aggregate Flow Fairness in MANs

Abstract

MANs and backbone networks are shared by "users" that can be individuals, organizations as well as communication service providers. Such users produce concurrently a variety of traffic patterns from multiple network locations. Traditional fairness definitions allocate bandwidth to individual source to destination flows. These individual flow fairness definitions generally allocate more bandwidth to users with a larger number of flows, thus creating unfairness at the user level. This paper explores alternative user level fairness criteria. We examine several extensions of the max-min fairness that allocate bandwidth fairly to users. We require the new criteria to be based on a max-min definition, to be backward compatible with the traditional max-min fairness (when each user has a single flow) and not to allocate zero bandwidth to any individual flow. We describe three different criteria for fair bandwidth allocation to users. The first is a weighted max-min criteria, achieving user fairness by the weights assigned to each flow; The second attempts to balance the user allocation separately over each link; Finally, we introduce a novel scheme termed the redefined vector-space fairness that is based on a lexicographical maximization of both user and individual flows. This paper evaluates the three fairness definitions both behaviorally and numerically. The simulation results show a clear advantage for the redefined vector-space fairness both in terms of user fairness and overall throughput.