基于网络流量预测的DASH系统优化  

Optimization of DASH System Based on Network Traffic Prediction

作　　者：耿俊杰 李晓明 颜金尧 GENG Junjie;LI Xiaoming;YAN Jinyao(Collaborative Innovation Center,Communication University of China,Beijing 100024,China;Beijing Thunisoft Information Technology Co.,Ltd.,Beijing 100024,China)

机构地区：[1]中国传媒大学协同创新中心,北京100024 [2]北京华宇信息技术有限公司,北京100024

出　　处：《计算机工程》2021年第5期292-300,共9页Computer Engineering

基　　金：国家自然科学基金(61971382);国家重点研发计划(2019YFB1804300)。

摘　　要：近年来基于超文本传输协议(HTTP)的自适应视频流量大幅上升,传统HTTP动态自适应流(DASH)速率算法无法准确预测网络吞吐量,导致网络带宽波动,使传输控制协议慢启动并触发抛弃规则,从而降低视频质量。提出一种基于网络流量预测的改进DASH速率算法。将DASH算法分为视频质量选择阶段、视频下载阶段和请求等待阶段,在视频质量选择阶段引入支持向量回归模型和长短期记忆网络预测网络吞吐量,结合缓冲时长选择更优质量的视频片段,在视频下载阶段通过预测实时吞吐量降低触发抛弃规则的次数。仿真结果表明,该算法可自适应流速率并减少抛弃规则的命中次数,有效提高视频体验质量。In recent years,the adaptive video traffic based on Hyper Text Transfer Protocol(HTTP)has increased greatly.The traditional Dynamic Adaptive Streaming over HTTP(DASH)rate algorithm can no longer accurately predict the network throughput,giving rise to the fluctuations of network bandwidth.The fluctuations cause the transmission control protocol slow to start and thus trigger the discarding rules,leading to a decrease in the video quality.This paper proposes an improved DASH rate algorithm based on network traffic prediction.The DASH algorithm is divided into video quality selection stage,video downloading stage and request waiting stage.In the video quality selection stage,the Support Vector Regression(SVR)model and the Long Short-Term Memory(LSTM)network are introduced to predict the network throughput.In addition,the buffer duration is used to select the better quality video clips.In the video downloading stage,the real-time throughput is predicted to reduce the number of times the discarding rules are triggered.Simulation results show that the proposed algorithm can adapt to the streaming rate and reduce the hit times of discarding rules,effectively improving the Quality of Experience(QoE)for video.

关 键 词：基于HTTP的动态自适应流 体验质量 吞吐量预测 支持向量回归模型 长短期记忆网络 

分 类 号：TP368.6[自动化与计算机技术—计算机系统结构]