Recently, mobile data traffic has been explosively increased and it is expected that the network capacity will be insufficient to accommodate all of the data traffic, and hence, the importance of quality-of-service (QoS) guarantee has also increased. In the communication systems, QoS is estimated by measuring several aspects of the network such as service availability, throughput, bit error rate, delay and jitter (delay variation) depending on the applications. Such aspects are largely divided into throughput related aspects and delay related aspects. Thus, we study the scheduling problems to enhance QoS of users in terms of throughput and delay by assuming that the network capacity is determined as follows.
First, we study the scheduling problem for achieving the best delay performance in the wireless network where the traffic arrivals lie interior to network capacity region. We analyze the individual delays and convergence performance under a generalization of the Max-Weight Scheduling (MWS) policy called $MWS- \omega$ policy, which is known to be throughput optimal. Based on this, we characterize the $MWS- \omega$ policies to achieve min-max fairness of individual delays and convergence time minimization.
Next, we propose a unified scheduling policy to achieve diverse QoS requirements in the wireless network. We consider the scheduling problem for achieving throughput QoS (weighted proportional fairness and minimum data rate) and delay QoS (maximum delay bound). Specifically, we define the throughput-utility function to guarantee throughput QoS, and develop the flow control and scheduling policy that maximizes the throughput-utility while guaranteeing delay QoS.
Finally, we study the scheduling problem for finding the work-conserving policies that achieve the best tradeoff between efficiency and fairness in terms of average flow delays in the data center networks. Specifically, we develop the flow scheduling policy that minimizes $L_p$ -norms of average flow delays by fully utilizing the dynamic information of remaining processing time of flows and starvation of users.