【作者】 杨鹏；

【导师】 喻莉；

【作者基本信息】 华中科技大学 ， 信息与通信工程， 2018， 博士

【摘要】 近年来,随着移动终端数量爆发式增长和新型移动应用的涌现,敏捷高效地响应海量移动用户请求面临着巨大的挑战。当前流行的基于云计算的移动应用运营模式的缺陷逐步凸显:由云服务器处传输海量数据至用户端对网络链路带宽造成了巨大压力,增加了响应用户请求的时间,显著影响了用户体验。在此背景下,新型的移动边缘计算架构被提出,具有计算和存储资源的边缘节点被大量部署在接入网内,以提升用户设备承载移动应用能力并降低响应时间,缓解无线回程网络带宽压力。由于高清视频和图像等多媒体流应用内容丰富、表现力生动,其在无线网络中的流量占比不断攀升。在移动边缘网络架构下,利用边缘节点实现高效的多媒体内容分发对提升移动用户体验和增加边缘系统资源利用效率具有重要意义。为此,本文从如下三个方面对边缘系统的多媒体内容分发和采集方法展开研究:移动边缘网络架构设计与多维资源分配联合优化、边缘节点多媒体内容存储策略和基于边缘节点地理位置的多媒体内容采集方法。最后,本文将归纳总结出相关领域研究空白和对未来工作的展望。首先,本文将考虑当前移动应用的多样化需求及边缘节点资源特征,设计边缘-云节点互协作框架。通过层次化的边缘结构和中心化的请求处理方式,该框架能在边缘-云协同作用下对用户请求处理作出最优决策,具有高效响应用户请求的能力。随后,基于该协作框架,本文研究边缘系统服务各类多媒体应用时的多维资源分配问题。相较于云服务器,边缘节点数目更多但节点各类资源有限。为提高移动多媒体用户体验和边缘资源的利用率,边缘系统内的存储、计算和通信资源需要被合理分配与优化调度。针对该问题,本文建立了一种基于网络状态和流媒体分发应用特征的资源需求模型。以优化各类流媒体分发质量为目标,本文提出了一种基于强化学习的多维资源分配算法,在理论上具有逼近后验最优算法的性能。其次,本文将研究基于边缘节点位置特征的内容存储算法设计。由于多媒体文件数据量巨大,而边缘节点存储资源有限,因而需要对各多媒体文件进行流行度预测和分析,以将流行度较高的媒体内容缓存在本地。高效的存储算法可以有效增加本地处理的媒体内容请求量,降低处理用户请求的延时,提高边缘节点资源利用率,并降低无线网络回程带宽压力。为实现上述目标,本文提出基于边缘节点的计算与存储资源,对用户请求和多媒体内容特征进行精细化的数据采集与预测分析,进而制定定制化的内容存储策略,以提高边缘节点本地响应的用户请求量。由于边缘节点位置特征具有多维特性,其对内容请求量的影响未知且随时间变化,为了制定精细化的存储策略,需对此类特征进行持续估计。在此背景下,本文将首先建立用户请求与地理位置特征之间的数学模型。在内容的边缘节点特征已知的条件下,建立并分析边缘节点位置特征向量与用户请求的关系。基于上述关系模型,本文将设计高准确度的地理位置特征在线预测算法。基于历史数据,通过强化学习的方式不断对边缘节点特征进行在线预测。最后,根据上述位置特征在线预测算法和用户请求数学模型,设计相应的边缘存储算法,并对其进行理论和数值分析,并获取实际数据对所设计的边缘存储算法进行验证。然后,我们将研究基于边缘节点空间位置信息的多媒体内容众包采集。传统众包采集方法主要依赖于云平台,而基于地理位置信息的边缘节点内容采集具有更广泛的前景。地理位置相关的信息可以直接在边缘平台采集并处理,进而有效降低传输至云平台处的数据量。基于边缘节点的上述优势,本文提出在边缘节点采取平台即服务(Platform as a Service,PaaS)的方式,探索高效的边缘空间内容众包采集方法。具体地,本文将首先设计基于边缘平台的内容众包采集模式。在众包服务背景下,内容需求方首先向众包平台提出内容采集请求,并提供相应的奖励机制。在此众包模式下,本文将设计空间内容众包采集的奖励机制以及参与者选择算法,利用边缘节点计算和存储资源通过强化学习的方式,动态选取众包应用参与者。该长效机制和在线算法有效规避了参与者的自私行为,增加了任务请求者的效用,并实现了众包平台长期累积效用的最大化。最后,本文通过理论分析对该算法相比后验最优策略的效用损失进行量化;通过抓取实际众包应用数据,对所设计算法的性能进行实验验证。最后,本文对以上研究内容和创新点进行了总结,基于移动边缘网络架构发展趋势和多媒体内容存储与传输领域的研究动态对本文的扩展工作进行了展望。作为移动边缘网络架构和多媒体内容存储、传输的交叉研究方向,本文对多媒体内容在边缘网络中的分发和采集为移动边缘系统支撑新型多媒体应用提供了理论依据、性能分析方法和实际部署方案。更多还原

【Abstract】 Recently,with the increasing of mobile devices and the explosion of mobile applications,providing agile and efficient responses to the surging mobile user requests is becoming challenging.Specially,the drawbacks of the cloud-based computing paradigm become apparent:getting massive data into and out of the cloud wirelessly requires substantial spectrum resources and incurs unexpected latency,which lead to degraded quality of experience.To address those challenges,the paradigm of mobile edge computing has been proposed,edge nodes with excessive computing and storage resources are massively deployed within the radio access networks.Edge nodes are able to process user request immediately,which significantly boosts the capacity of mobile users,reduces the response time of user requests,and effectively alleviates the bandwidth pressure of the wireless backhaul.In mobile network,multimedia contents(e.g.,high-resolution video and image)are gaining increasing popularity due to their rich information and vivid expression.As a result,the proportion of multimedia traffic,is projected to be more than 80%of total wireless traffic in 2020.In the new paradigm of mobile edge computing,efficiently disseminate and collect multimedia content is non-trivial in improving user experience and the effectiveness of edge resources.To this end,this thesis is committed to the following three aspects in multimedia content dissemination and gathering:design of mobile edge computing architecture and analysis for multi-dimensional edge resource allocation,high-efficient mobile edge caching policy and edge location-based spatial content crowdsourcing.In the end,the contribution and novelty of this work will be summarized,and potential research issues and future work will be presented.Firstly,considering the diversified requirements of mobile applications and the characteristics of edge resources,we design an edge-cloud interoperation framework.Through hierarchical architecture and centralized management,this framework is able to make better decisions to process user requests,providing users with rapid responses.Based on this framework,we then investigate the allocation strategy of the multi-dimensional edge resource.Compared with cloud platform,the amount of computing and storage resources at each edge node is limited.To improve user experience and the effectiveness of edge resource,the usage of edge resources should be well managed.To address this issue,we propose a resource demand model for multimedia content distribution,taking into account user preferences and system states.To improve user experience,a reinforcement learning based multi-dimensional resource allocation algorithm is proposed.Theoretical analysis indicates that the performance of the proposed algorithm approaches that of the hindsight optimal.Secondly,to reduce the latency of multimedia content dissemination,we investigate edge content caching algorithms.Since the volume of media content is often higher than that of other content types,to fully exploit the limited edge storage resources,only the most popular contents can be cached locally.Efficient caching algorithms will significantly increase the content hit rate,and hence reduce the latency of content requests.To this end,we propose to conduct dedicated data gathering and analysis on user requests and content feature based on edge computing and storage resources.Observing that user demands on certain contents vary greatly at different locations,we devise location-customized caching schemes to maximize the total content hit rate.Specifically,a linear model is used to estimate the future content hit rate.For the case where the model noise is zero-mean,a ridge regression based online algorithm with positive perturbation is proposed.Regret analysis indicates that the proposed algorithm asymptotically approaches the optimal caching strategy in the long run.When the noise structure is unknown,an H∞filter based online algorithm is further proposed by taking a prescribed threshold as input,which guarantees prediction accuracy even under the worst-case noise process.Both online algorithms require no training phases,and hence are robust to the time-varying user demands.The underlying causes of estimation errors of both algorithms are numerically analyzed.Moreover,extensive experiments on real world dataset are conducted to validate the applicability of the proposed algorithms.It is demonstrated that those algorithms can be applied to scenarios with different noise features,and are able to make adaptive caching decisions,achieving content hit rate that is comparable to that via the hindsight optimal strategy.Then,we investigate multimedia content gathering in the form of spatial crowdsourcing based on edge platform.Most existing crowdsourcing platforms are operated on cloud server.With the immersive deployment of edge resource,location-based service,such as spatial crowdsourcing,can effectively improve of way of multimedia content gathering.Human workers in a specific location are harnessed to collect geographical data,after which the data are timely processed on the edge platform,which further reduces the amount of traffic delivered to the cloud.To achieve better performance,spatial crowdsourcing task relies on both worker’s effort and skill.Therefore,to maximize the long-term platform utility,we exploit edge platform as a service to identify valuable workers through learning their performance information.Worker’s historical performance data are recorded at local fog server,based on which valuable workers are identified and selected to perform the media content gathering tasks.During worker selection,we aim at balancing the exploration and exploitation,and propose an online algorithm that promotes workers who are not fully explored.With budget constraint,the proposed algorithm is able to maximize the long-term platform utility for content gathering.Theoretical analysis indicates that the proposed learning algorithm achieves asymptotically diminishing regret.Finally,extensive simulations on real-world dataset are conducted,which demonstrate the advantage of our algorithm over other methods.Finally,the contribution and novelty of this thesis are summarized.Based on current trends of mobile edge network architecture and the caching and distribution of multimedia content,potential research issues and future work are presented.As an interdisciplinary research topic,the work of edge-based multimedia content dissemination and gathering presented in this thesis provides fundamental principles,performance analysis and practical deployment solutions.更多还原