| Recent years have witnessed the increasing popularity of online social networks andonline video sharing networks, as well as the rapid convergence of the two networks.Reshaping the video distribution landscape, delivery of online social video contents hasbecome an intriguing research direction. Traditional content delivery approaches that aredesigned without consideration of social topology and user behavior, are less efficient ifnot completely ineffective for social video delivery. To address the social video deliveryproblem, by jointly considering the social information, user behavior, and content fea-tures, this thesis focuses on charactering the propagation of social videos, the networktopology and resource allocation for social video delivery, propagation-based content de-ployment, and social video service deployment. Contributions of this thesis are summa-rized as follows:1. Based on large-scale measurement studies, we reveal characteristics of the socialvideo content delivery. These characteristics include the impact of social relation-ship and user behavior on content propagation, the edge-to-edge content collectionand distribution patterns, the flatted popularity distribution of social video contents,the heterogeneous user preference, and the dynamics and localities in content prop-agation. We also present analysis on the inefficiency of traditional content deliveryparadigms, and summarize guidelines for designing a social-aware content deliveryframework.2. We propose a social-aware delivery architecture and the corresponding resource(e.g., bandwidth) allocation strategies. An edge-oriented and geo-distributed archi-tecture is proposed for social video delivery, so that edge users can benefit fromdownloading from the geo-scaling servers. We design strategies to partition usersso that not only can users download from their preferred servers but also is thereplication cost for the system minimized. Meanwhile, edge server bandwidthsare allocated according to the prediction of video propagation, for an efficient andproactive resource utilization.3. We further propose propagation-based content deployment strategies. Contentsare replicated according to the propagation patterns, including social locality, geo-graphical locality and temporal locality. Predictors are designed to predict where a content will propagate to, how many global users it will attract in the propaga-tion, and how many local friends of a user it will attract. These predictions willguide how contents are replicated across the edge-cloud servers and cached by lo-calpeers,suchthatuserscandownloadcontentsfromserversorpeersclosetothem,achieving good quality-of-experience (QoE).4. To provide a uniform platform for content processing and delivery in social videoservices,weproposeahybridcloud-basedsocialvideoservicedeploymentstrategy,inwhichcontentcollection,contentprocessing,andcontentdeliveryarejointlycon-sidered. According to the different resource requirement in content processing andcontent delivery, local processing using IaaS (Infrastructure as A Service)-basedcloud instances and global distribution using PaaS (Platform as A Service)-basedcloud platform are connected by content propagation predictions. Optimizationsare formulated to solve the cloud resource allocation and content replication prob-lems,anddistributedalgorithmsaredesignedforreal-worldsocialvideoapplicationdeployments. Our strategy can significantly reduce the content processing delay. |
PDF Full Text Request