A Scheduling Framework for Adaptive Video Delivery over Cellular Networks Ashwin Wuluwarana University of Texas at Dallas [email protected] Preetam Mallappa University of Texas at Dallas [email protected] 1. INTRODUCTION With the increasing consumer demand for mobile video streaming [1] and to enhance the Quality of Experience (QoE), the need for new techniques to effectively utilize the available resources is increasing. One such technique is Dynamic Adaptive Video Streaming over HTTP (DASH) [2] which is quickly getting adopted as the technology of choice for the delivery of video traffic over internet. It is emerging as a promising method to deliver video to mobile users with enhanced QoE. At the same time, the popularity of Internet-based high-quality streaming for various end-user devices such as HDTVs, mobile phones, gaming devices, and computers is on a steady rise as well. The bandwidth requirement for such devices is rapidly increasing as the content quality is improving to meet end-user demands. With abundant video content and increasing bandwidth demands, it is becoming likely that two or more adaptive streaming players have to share a network bottleneck and compete for available bandwidth. And studies in the past [3,4] confirm that the video commercial players’ performances degrades when multiple DASH flows share a bottleneck link on the internet. In this respect, there are three potentially conflicting goals that a robust adaptive video algorithm must strive to achieve: Fairness: Multiple competing players sharing a bottleneck link should be able to converge to an equitable allocation of the network resources. Efficiency: A group of players must choose the highest feasible set of bitrates to maximize the user experience. Stability: A player should avoid needless bit-rate switches as this can adversely affect the user experience. In particular, according to the studies [4] it confirms that when two players compete at a bottleneck link, these players fail to meet one or more of the above mentioned goals. Therefore, there is a high demand for multi-user DASH optimization techniques. One other fundamental problem with DASH flows is the inaccurate estimation of network bandwidth. From the evaluation study presented in [5], they concluded that network level and radio-level adaptation is required for enhancing service capacity and user perceived quality. Also recently, Authors in [6] have proposed a novel scheduler which they name as Adaptive Guaranteed Bit Rate (AGBR) scheduler, that optimally and adaptively sets the Guaranteed Bit Rate (GBR) for each video flow in LTE network with heterogeneous traffic. The approach is intended to achieve a level of fairness among the video flows while preventing starvation of other data flows, but in this framework the definition of utilities are not content-aware which might not lead to the best possible Quality of Service (QoS) among multiple video flows. Hence there is a need for a resource management framework that addresses the above challenges effectively. In summary, through this paper we will make the following contributions: We systematically explore the design space for optimization of DASH algorithms in multiple DASH flows over shared network environment, together with satisfying the goals of fairness, stability and efficiency. We identify the major factors in bit-rate selection and scheduling mechanisms employed that influence the DASH players to cause instability. We also recognize various mechanisms for flow scheduling, bandwidth estimation, and bit-rate selection that impact the design of adaptation algorithms, thereby addressing the tradeoff between stability, fairness and efficiency. 2. MOTIVATION/PROBLEM STATEMENT Although the current generation cellular base stations utilize advanced radio resource management techniques for single-rate video streaming and elastic data flows over multiple users, the system still lacks intuitive structure to handle adaptive video streaming services across multiple mobile users. It becomes more challenging when multiple adaptive videos concurrently compete for bandwidth over a shared network. This competition can potentially lead to three performance issues, which are video player instability, unfairness among video players, and inefficient utilization of bandwidth.[3,7,8]. In allocating resources for adaptive video traffic, there arises conflicting situations. To achieve expected Quality Of Experience(QOE), significant bit-rate switches for users becomes inevitable. Hence an effective framework that achieves an optimal balance among the two is required. Through this paper, we intend to address these issues by exploring existing designs and thereby suggesting possible areas of refinements. When it comes to delivering video over any network, it becomes imperative to meet certain performance expectations. Challenges add up when it comes to streaming adaptive video over cellular networks. These additional challenges to be satisfied are (a) optimal allocation as desired by the cellular operator (b) bit-rates stability allocated to a user and (c) maintaining high resource utilization. Additionally, a recent study [9] confirms that most commercial video players fail to meet the above mentioned goals due to inaccurate estimation of the network bandwidth. This inaccuracy leads to under-estimation or over-estimation of the underlying bandwidthwhich is caused due to temporal overlap of the chunks of different players when they periodically download and estimate per-chunk throughput. The current flow schedulers in LTE systems are not configured to exploit the variable sized video chunks, which forms the basis for adaptive video delivery. Hence, the network operators are unable to optimally predict the allocation of resources among users who stream adaptive videos. Additionally, current schedulers cannot handle bit-rate switches. And also, while fetching adaptive video chunks, it requires the players to download small-sized chunks periodically and estimate the network bandwidth, for which the present schedulers are
View Full Document