Predictive Two-Timescale Resource Allocation for VoD Services in Fast Moving Scenarios

Abstract

With the burgeoning video applications in 5G/B5G, the requirements for video frame quality and playback smoothness are equally strict but challenging to satisfy simultaneously, especially in fast moving scenarios. To tackle this issue, in this paper, we propose a predictive two-timescale resource allocation scheme for video-on-demand (VoD) services by leveraging the long-term channel prediction. The scheme addresses two critical concerns. One is how to schedule packets over a large timescale to avoid transmission in poor channel states. The other is how to guarantee the delay-quality of service (QoS) over a small timescale to satisfy the high-quality requirements of video services. In the framework of network slicing, we split the VoD slice into two logical sub-slices to support the video contents that play immediately after downloading and the video contents to be pre-cached, respectively. To perform an efficient resource reservation, we propose a martingales-based resource estimation method for the video streams with statistical delay-QoS requirements. Based on this, our scheme is divided into two stages. First, on the time scale of prediction time slots, we propose a low-complexity heuristic algorithm to find a spectrum-efficient video delivery pattern. Then, in each transmission time interval, we develop a utility theory based resource allocation algorithm to balance the metrics of spectrum efficiency, fairness, and delay-QoS. The simulation results demonstrate the capability of the QoS guarantee and the promising gain of spectrum efficiency brought by our scheme.