Measurement-based scheduler for multi-class QoE optimization in wireless networks

摘要

Traditional wireless schedulers have been driven by rate-based criteria, e.g., utility maximizing/proportionally fair, and/or queue-based packet schedulers which do not directly reflect the Quality of Experience (QoE) associated with flow-based transactions and services. This paper proposes, a Measurement-Based Delay Optimal (MBDO) scheduler, which optimizes a cost function of the mean flow delays in a multi-class system, e.g,. web interactive, file downloads, etc. In this context the cost function expresses desired trade-offs amongst traffic classes reflecting heterogeneous QoE sensitivities which are nonlinear in the flow delays and/or system loads. To achieve optimality, MBDO scheduling uses measured system variables and knowledge (or measurement) of class flow-size distributions to adapt a weighted Gittins index scheduler. We show that under mild assumptions, and in a stationary regime, that MBDO scheduling is indeed asymptotically optimal. Perhaps more importantly, MBDO schedulers can self-optimize by adapting to slowly varying traffic loads, mixes and flow size distributions. Our extensive simulations confirm the effectiveness at realizing trade-offs and performance of the proposed approach.