【作者】 刘强；

【导师】 彭凯；

【作者基本信息】 华中科技大学 ， 信息与通信工程， 2019， 硕士

【摘要】 随着智能移动设备使用的增长,互联网流量正以惊人的速度飞速增长,其中视频数据逐渐成为互联网数据中的主流数据形式。海量数据为云供应商和视频内容提供商带来了巨大的压力。移动边缘计算(Mobile Edge Computing,MEC)作为一种经济有效的模式出现,在网络边缘靠近用户的地方提供存储、计算和网络资源。视频内容提供商能够利用MEC提供的资源部署低延迟和计算密集型视频服务。如何在满足服务质量(Quality of Service,QoS)的前提下尽可能节约MEC资源的租赁成本,成为视频内容提供商关心的重要研究课题。本文提出了一种联合协同缓存、处理和资源租赁模型,旨在帮助视频内容提供商在视频变体缓存放置方案、视频请求路由方案和资源租赁方案上做出更好的决策,最终使得MEC资源的租赁代价最小。为适应用户需求的多样性,模型考虑了自适应比特率视频流技术,这使得用户能够根据自身的网络状况、移动设备的处理能力和自身喜好调整对某一特定比特率版本视频的需求。为更加贴近现实世界的场景,模型考虑了带宽限制条件,由于回程链路的网络资源有限,带宽更能成为视频传输的瓶颈。模型的NP完全性使得不存在对原始问题的非平凡解,为问题的求解造成了困难。受到“分治”思想的启发,本文将原始的联合建模问题分解成两个子问题:(i)利用当前MEC系统内未逾期的MEC资源最大化能够覆盖的用户请求数目和(ii)最小化新租赁的资源代价同时能够覆盖当前MEC系统内未处理的用户请求。对于子问题(i),本文将该子问题转化为单调次模函数的最大化优化问题,并给出了基于自适应比特率的主动式缓存算法,基于求解得到的缓存方案本文给出了近似在线请求路由算法。对于子问题(ii),本文利用拉格朗日松弛方法对原始问题中较复杂的限制条件进行松弛,将子问题(ii)分解为两个子问题,最后利用次梯度优化算法对子问题的解进行调整。本文通过实验仿真验证了算法在提升缓存命中率、减少根服务器到MEC系统的回程链路负载和最小化租赁代价方面的价值,并和传统方法进行了对比。更多还原

【Abstract】 With the rapid growth of adoption of smart mobile devices,the internet traffic is now increasing at an amazingly fast speed,among which video data has gradually become the mainstream data.The massive data volume brings great pressure to cloud vendors as well as video content providers.Mobile-Edge Computing(MEC)shows up as a cost-effective paradigm which provides storage,computation and network resources in proximity to users residing at the network edge.Video content providers are able to deploy low-latency and computation-intensive video services thanks to the resources that MEC provides.Thus how to save MEC resource leasing cost as much as possible while still satisfying the Quality of Service(QoS)becomes an important topic to video content providers.In this paper,a joint collaborative caching,processing and resource leasing model is proposed to help video content providers make better decisions on the video variants placement,video request scheduling and resource leasing strategy so as to minimize the leasing cost of MEC resources.The model takes Adaptive Bitrate video streaming into consideration to better meets the diversity of user’s requests,which allows users to adjust their needs for some specific bitrate versions of videos according to their network conditions,mobile device capacity and their preferences.Bandwidth constraints are also considered to provide a simulative scenario more close to the real world scenario,which indeed could be the bottleneck of video transferring because of the limited network resource of backhaul link.The NP-complete property of the model sets great barriers to solve the primal problem as no non-trivial algorithms for this model existed.Inspired by the thoughts of ‘divide and conquer’,this paper decomposes the primal problem into two subproblems:(i)maximizing the number of the users’ requests covered using the unexpired sources remained in the current MEC system and(ii)minimizing the resource leasing cost while meets all the unhandled requests in the current MEC system.For subproblem(i),this paper transforms the subproblem into a monotone submodular maximization problem and gives an ABR-aware proactive caching algorithm and an approximately optimal online request routing algorithm based on the given caching strategy.For subproblem(ii),this paper incures the Lagrangian Relaxation method to relax the complex constraints in the subproblem and decomposes it into two less complex problems.Subgradient algorithm is used to adjust the solutions given by solving the above-mentioned two less complex problems to provide an approximately optimal solution to subproblem(ii).Massive simulations was taken to prove the merits of our algorithms in increasing cache hit ratio,reducing the backhaul traffic load between root servers and MEC systems and minimizing resource leasing cost compared to conventional approaches.更多还原