Max-min fairness

Max-min fairness is a criterion for dividing scarce resources, mainly in communication networks, that is met when an allocation is feasible and any attempt to raise one participant's share would necessarily lower the share of another participant who already has an equal or smaller allocation. Compared with first-come first-served scheduling it shapes traffic, because an ill-behaved flow of large packets or bursts mostly penalizes itself and not other flows. Fair queuing is an example of a max-min fair packet scheduling algorithm, and round-robin scheduling is max-min fair when packets are equally sized. For capacity that is allocated in advance, the allocation can be found by progressive filling, which grows all rates together from zero until some link saturates, freezes the flows crossing it, and continues with the rest. A max-min fair allocation, where it exists, is unique, and it trades some average throughput for more stable service quality than a maximum throughput policy gives.