| Crowdsourced live streaming becomes more and more popular on the Internet in recent years.In such a video streaming system,it is challenging to allocate limited service resources,including the transcoding and delivery resources,among the numer-ous broadcasting channels.Existing solutions of studies and applications only consider channels' popularity in resource delivery(e.g.,Twitch)-But unlike the platforms over-seas,the domestic crowdsourced live streaming platforms like Douyu enable viewers to perform rich interactions,such as posting comments and donating monetary virtual gifts,through the danmu mechanism in broadcasting channels.That' s why we call them "Danmu-Interactive crowdsourced live streaming platforms".As a consequence,channels will have new properties other than popularity that are derived from the danmu interactions.Therefore,in order to balance channels' popularity with their profitability,we propose that danmu interaction derived channel properties should be considered in resource allocation.In this paper,we use Douyu danmu-Interactive live streaming platform as an exam-ple.By performing a measurement study on Douyu,we find that its system is deployed upon a video delivery network(VDN),and the same as Twitch,it allocates service re-sources based on channels' popularity.By analyzing Douyu danmu communication protocol and using the third-party developper API,we collect a large scale danmu mes-sage dataset from Douyu.Then we characterize the danmu interactions,analyze the viewers' behavior and broadcasters' behavior and profile the interaction derived chan-nel properties.We find that the properties are highly skewed and heavy tailed.By using these properties,we propose improved algorithms for allocating transcoding re-sources and strategy for allocating delivery resources.Real-world trace driven simu-lation shows that our approach can effectively improve the overall performance of the danmu-interactive crowdsourced live streaming systems. |
PDF Full Text Request