【作者】 王超；

【导师】 龙承念；

【作者基本信息】 上海交通大学 ， 自动化， 2016， 硕士

【摘要】 随着3G/4G移动网络技术的飞速发展,智能设备全球性普及,移动互联网正以前所未有的速度颠覆人们生活。然而,电池发展没有突破技术瓶颈,难以长时间维持智能设备大量应用运行,待机时间远远不够。考虑到网络通信耗能占终端能耗主要组成部分,终端移动网络通信能耗优化问题成为移动互联网领域研究热点。其中存在的尾部效应是导致终端通信能效较低的主要原因,因为每次无线接口数据传输结束,接口不能立即转换到低功耗状态,而是维持高功耗状态一段时间,以便快速响应随后的网络传输。这段等待的时间则是尾部时间。减少尾部效应是挑战性难题。尾部机制保证接口在高功耗状态快速响应用户请求,保证用户实时性体验。若直接简单去除尾部时间,新的传输到达时,接口需要从原来的低功耗状态切换到高功耗状态,需要向基站申请链路资源分配,同样费时费力。因此需要在能效与用户体验之间作出取舍。考虑到网络数据主要来源于应用,因此研究应用网络数据的传输模式,掌握其在时间轴的分布特性,作为接口进行高低功耗之间转换的参考,根据应用类型或者流量模式动态决定尾部时间,改变原来应用数据传输与接口之间缺乏信息交互的模式,是有效的解决方案。本文正是基于这样的研究思路,通过收集20名用户持续两周的终端联网使用数据,深入研究流媒体、网页浏览、新闻阅读等应用的流量模式,发现应用存在周期性同服务器进行通信的现象,严重消耗大量能量与无线资源。同时,通过对数据流传输间隔时间(IBT)信息熵计算,发现应用数据流的传输序列IBT存在很强的时间相关性。基于此我们提出高能效终端无线通信能耗优化机制-TATO。该设计主要包括数据流的监测,间隔时间预测,快速休眠触发。TATO通过流量监测器获取终端流量信息,从中提出数据传输间隔序列(IBT)以及传输端口、传输协议等相关网络特征,同时利用SVM模型对数据流的间隔时间进行建模,根据时间相关性特征预测之后的间隔时间是否超过阈值,从而主动调整接口传输状态,在不影响传输的前提下,减少尾部时间。TATO主要优势在于:TATO对正常的数据传输影响较小,对用户体验影响较小。其次,TATO是轻量级的解决方案,作为后台应用运行于终端,无需进行系统层面修改或者对其他应用作出要求,非常便于开发。通过trace-driven仿真实验,证实TATO是有效可行的,结果表明TATO能够节省50%-60%能量,而且不会影响用户体验。更多还原

【Abstract】 With the rapid development of 3G/4G mobile network technology and worldwide popularity of intelligent devices,mobile Internet is growing at an unprecedented rate to subvert the people’s lives.Significantly enhance of computing performance makes a wide variety of applications such as streaming media players,VOIP,video conference,etc.run in smart devices fluently,which greatly enriched and changed the way people live.They have become an indispensable part of life.At the same time,the development of battery technology did not break the bottleneck and it is difficult to maintain a large number of applications running for a period time.Taking into account the network communication energy consumption accounted for the major part of the mobile terminal,energy optimization of mobile Internet has become a hot area of research.The presence of the tail effect is the main reason of low terminal communication efficiency because after data transmission completion,interface does not immediately switch to IDLE state but to maintain the high power state for a period of time in order to quickly response the subsequent network transmission.This waiting time is the tail of energy during which there is no data transmission.Articles [2],[13] pointed out that this part of this energy accounted for nearly 60 % of the total energy consumption of the interface.Reducing tail effect is challenging problem.The tail mechanism is designed to achieve the quickly response to upcoming data transmission.For energy optimization,we need consider the tradeoff between energy saving and promotion overhead which needs several message exchanges between terminal and base station and consumes CPU computation resource,causes time delay.Considering the data request come from applications,it is better to study the traffic pattern of application and search the distribution in time so that interface can dynamically adjust RRC state according to current data transmission pattern.In this paper,we collected the traffic trace of 3G/4G from 15 users.According to analysis of trace from different applications such as web browser,new reading,streaming and so on,we discovered the period transfer and the impact on energy consumption and radio resource of terminal.As for this problem,we propose period detect algorithm to distinguish these transmission and reshape traffic for energy saving.Further,for traffic pattern learning,we build model for data streaming from applications and verified that the data transmission has strong temporal correlation.Based on observation,we design high-efficiency energy optimization scheme,called TATO.TATO is mainly consist of three parts.The core idea is to train SVM model to learn traffic model,based on which TATO predicts arrive time of following data transmission within predefined time.With the predict result,TATO adopts the fast dormancy mechanism to adjust RRC state to cut unnecessary tail time to achieve energy saving without much influence on user perspective experience.The advantage of TATO is obvious: it accomplish significant energy saving with little influence on application running;moreover,it is light-weight and runs in application level without requirement to system and other application.We implemented trace-driven simulation based on trace and result shows it is feasible and can saves 50%-60% energy on average across different applications.更多还原