On Crowdsourced Live Video Streaming with Heterogeneous Broadcasters and Viewers

Recent years have witnessed the booming popularity of Crowdsourced Live Streaming (CLS) platforms, through which numerous amateur broadcasters lively stream their video contents to viewers around the world. The heterogeneous quality and format of source stream however require massive computational resource to transcode it into multiple industrial standard quality versions to serve viewers with distinct configurations. In this thesis, we analyze the large dataset we captured from the popular CLS platform Twitch TV. We then present a generic framework utilizing the powerful and elastic cloud computing services for crowdsourced live streaming with heterogeneous broadcasters and viewers. We jointly consider the viewer satisfaction and the service availability/pricing of geo-distributed cloud resources for transcoding. We first develop an optimal scheduler for allocating cloud instances with no regional constraints, and then extend the solution to accommodate regional constraints. However, given the considerable cost of cloud services, and the fact that the CLS platform charges nothing from viewers as a free system in nature, cloud-based transcoding solutions can only provide limited service, resulting in the current real-world situation. On the other hand, we witness huge computational resource among the massive fellow viewers in CLS systems that could potentially be used for transcoding. Inspired by the paradigm of Fog Computing, we propose CrowdTranscoding, a novel framework for CLS systems to smartly offload the transcoding assignment to the edge of network. We evaluate both of our novel frameworks with extensive trace-driven simulations and PlanetLab-based experiments, under diverse parameter settings. The superiority of our designs has been confirmed, while the experiment results also offer some further practical hints towards real-world migration.