On the Delay-Fairness through Scheduling for Wireless OFDMA Networks

Abstract

This paper studies QoS-guaranteed fair resource allocation and packet scheduling for OFDMA networks. We start with non-traffic-aware weighted generalized proportional fair (WGPF) scheduler and make it a delay-fair scheduler. Delay-fair objective is to equalize delay dissatisfaction measures among users. In this paper the focus is on mean queuing delay. We show how the WGPF objectives are connected and are equivalent to delay-fair scheduler, derived from Little's law. We see that our fair framework can also be interpreted as minimizing the total delay dissatisfaction that users are experiencing. This framework extends the conventional fairness notions in order to handle heterogeneity of traffic in time and among users which is an important emerging problem. We then study an important special case, which is min-max delay fairness. We prove that the developed framework for a special dissatisfaction function, asymptotically leads to the min-max average delays. The developed framework, in this special case, can be adjusted between two extreme objectives: minimizing total average delay among users and minimizing the maximum average delay among users. Finally we prove that the gradient scheduling algorithm is equivalent asymptotically to serving the user with the largest average delay at each iteration.