Empowered by today's rich tools for media generation and collaborative production, multimedia service paradigm is shifting from the conventional single source, to multi-source, to many sources, and now towards crowdsource, where the available media sources for the content of interest become highly diverse and scalable. Such crowdsourced livecast systems as Twitch.tv, YouTube Gaming, and Periscope enable a new generation of user-generated livecast systems, attracting an increasing number of viewers all over the world. Yet the sources are managed by unprofessional broadcasters, and often have limited computation capacities and dynamic network conditions. They can even join or leave at will, or crash at any time. In this thesis, we first conduct a systematic study on the existing crowdsourced livecast systems. We outline the inside architecture using both the crawled data and the captured traffic data from local broadcasters/viewers. We then reveal that a significant portion of the unpopular and dynamic broadcasters are consuming considerable system resources. Because cloud computing provides resizable, reliable, and scalable bandwidth and computational resources, which naturally becomes an effective solution to leverage heterogeneous and dynamic workloads. Yet, it is a challenge to utilize the resources from the cloud cost-effectively. We thus propose a cloud-assisted design to smartly ingest the sources and cooperatively utilize the resources from dedicated servers and public clouds. In current crowdsourced livecast systems, crowdsourced gamecasting is the most popular application, in which gamers lively broadcast game playthroughs to fellow viewers using their desktop, laptop, even mobile devices. These gamers' patterns, which instantly pilot the corresponding gamecastings and viewers' fixations, have not been explored by previous studies. Since mobile gamers and eSports gamers occupy a large portion of content generators. In this thesis, we target on two typical crowdsourced gamecasting scenarios, i.e., mobile gamecasting and eSports gamecasting, respectively. We investigate the gamers' patterns to explore their effects on viewers and employ intelligent approaches, e.g., learning-based techniques, to capture the associations between gamers' patterns and viewers' experiences. Then, we employ such associations to optimize the streaming transcoding and distribution.
Copyright is held by the author.
This thesis may be printed or downloaded for non-commercial research and scholarly purposes.
Supervisor or Senior Supervisor
Thesis advisor: Liu, Jiangchuan
Member of collection