Real-time QoE estimation of DASH-based mobile video applications through edge computing
Real-time QoE estimation of DASH-based mobile video applications through edge computing
复制标题
DOI：
10.1109/infcomw.2018.8406935
复制
发表时间：
2018-04
期刊：
影响因子：
--
通讯作者：
Chang Ge;Ning Wang
中科院分区：
文献类型：
--
作者：
Chang Ge;Ning Wang
近年来，使用MPEG-DASH（HTTP上的动态自适应流媒体，如YouTube和Netflix）的视频应用一直主导着互联网流量。越来越多的人认识到，为了向视频客户端提供更好的体验质量（QoE），内容服务提供商和网络运营商都需要首先了解客户端的QoE。在本文中，我们提出了一种新的实时QoE估计系统，通过边缘计算，已实现和部署在一个真实的LTE-A网络边缘。当配备有这样的系统时，部署在移动的网络中的任何虚拟网络功能（VNF）将能够真实的时间推断其覆盖范围内的所有DASH客户端的QoE，其中不需要来自客户端的反馈。此外，我们的计划是能够工作在忙碌网络环境中，包括空中接口，其中数据包错误频繁发生鲁棒性。这种方案的重要性在于通过在移动的边缘的非常轻量级的机制获得关于用户QoE的准确和实时的知识，其可以即时用于各种内容操纵或资源适配操作，以便确保动态条件下的用户QoE。通过在真实的LTE-A网络中的实验，我们证明了我们的方案能够以非常低的CPU和RAM占用量以非常高的精度估计DASH客户端的QoE。
Video applications using MPEG-DASH (Dynamic Adaptive Streaming over HTTP, such as YouTube and Netflix) have been dominating the Internet traffic in recent years. It is increasingly acknowledged that in order to provide video clients with better Quality-of-Experience (QoE), both content service providers and network operators need to be aware of clients' QoE in the first place. In this paper, we present a novel real-time QoE estimation system through edge computing, which has been implemented and deployed at a real LTE-A network edge. When equipped with such a system, any virtual network function (VNF) deployed in a mobile network will be able to infer all DASH clients' QoE under its coverage in real time, where no feedback from clients are needed. Furthermore, our scheme is able to work robustly in busy network environments involving air interface where packet errors frequently occur. The significance of such a scheme is the availability of accurate and real-time knowledge on user QoE through a very lightweight mechanism at the mobile edge, which can be instantaneously used for various content manipulation or resource adaptation operations in order to assure user QoE in dynamic conditions. Through experiments in a real LTE-A network, we demonstrate that our scheme is able to estimate DASH clients' QoE with very high accuracy with very low CPU and RAM footprint.