Font Size: a A A

Strategies Of Multimedia Content Delivery With Edge Network

Posted on:2018-05-22Degree:DoctorType:Dissertation
Country:ChinaCandidate:W HuFull Text:PDF
GTID:1368330566987977Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
Recently,the multimedia traffic is explosively increasing and has dominated the Internet traffic.The exponential growth of multimedia traffic far surpasses the speed of underlaying infrastructure improvement,making the traditional approach of deploying more servers ineffective.To satisfy requirement of efficient multimedia content delivery,more and more researchers rethink the current content delivery architecture and begin to pay attention to exploring novel delivery schemes.As a potential solution,edge network based content delivery network,which is constructed by the densely distributed edge devices,has been gaining lots of momentums.This solution becomes more promising with the improvement of edge devices capabilities.In this thesis,we study the distribution of edge resources and the pattern of users' content consuming,with a data-driven approach,to guide the key strategies involved in the edge-network based multimedia content delivery network.The main contributions of this paper are summarized as follows:·We propose Wi-Fi AP resources global scheduling and collaborative video replication strategies.By carrying out large-scale measurement studies,we first understandthe distribution of edge resources and users' request and users' content preference.These insights then provide valuable guidelines in the key strategies design,including on-demand resource deployment,global resource scheduling and collaborativecontent placement.By formulating the joint optimization of users' quality-of-experience and the operation cost as an optimization problem,we propose heuristicalgorithms to schedule the edge resource according to users' requests,reduce serverpeak load while improving users' quality-of-experience.·We propose Wi-Fi AP resources local scheduling and personalized video prefetchingalgorithms.In particular,two algorithms are designed for the TV series and commonvideo prefetching scenarios,respectively.The former utilizes the reinforcementlearning algorithm to learn the optimal prefetching actions by jointly consideringthe users' transition behavior and the server load dynamics,while the latter employs a tensor learning strategy to take both the videos' semantic information and the temporal pattern into account to predict users' affinity to different contents in the case with content popularity dynamic and data sparsity.Through simulation experiment,we prove that prefetching videos ahead of users' request can offload the peak server load while improving users' quality-of-experience(e.g.,the service latency reduction).·We propose request redirection-based dynamic content delivery strategies.By conducting measurement studies,we reveal users' webpage viewing pattern,including random failure occurrence,short online duration,small re-joint possibility and net work dynamics.Based on these characteristics,we estimate the network latency between browsers,predict remaining online duration and schedule users' requests. Our strategies utilize the WebRTC-powered browsers to construct an overly path to recover web content delivery path to improve the system stability.
Keywords/Search Tags:Content delivery network, Edge network, Resource scheduling, Quality of experience, Optimization
PDF Full Text Request
Related items