Generalized Fairness in QoS Scheduling

Many scheduling protocols have been developed that provide both rate and fairness guarantees. That is, flows can take advantage of unused capacity without incurring any penalty at the scheduler. The manner in which unused capacity is distributed among the active flows varies significantly between protocols. Some are partial to flows that have a larger reserved rate, while others are partial to flows with a lesser reserved rate. We present a generalization of all of these behaviors under a single common framework, which we refer to as generalized fairness. We present a scheduling protocol that provides generalized fairness, and we prove upper bounds on the exit times of packets from the scheduler. These bounds apply to earlier protocols as well, since they are specific instances of the general framework. In addition, the protocol also supports flows that transmit at a fixed rate but have tight delay requirements.