Please note: This seminar will be given online.
Sharon Choy, David R. Cheriton School of Computer Science, University of Waterloo
Live video streaming over wireless IP networks has increasingly replaced traditional solutions, such as using a television production van, for covering live events for the purpose of news broadcasting. The advantages of live video streaming over wireless include lower cost, no need for specialized infrastructure, and greater physical portability. Using multiple network interfaces can potentially improve video quality as we can aggregate available bandwidth or provide fault tolerance in the scenario where links are relatively independent. The main challenge of multi-homed video streaming is effectively using multiple, heterogeneous links to consistently deliver high-quality video while accounting for the end user’s preferences.
We present Conflux: a new protocol that models the multi-homed adaptive video bitrate problem as a decision making under uncertainty problem. Conflux determines a configuration, which is defined as the mapping of video data to links, that reflects the end user’s preferences. We present a new model where we decouple link characterization from link usage and video bitrate selection. Each link is empirically evaluated to estimate the probability that it is able to sustain a certain sending rate. Conflux statelessly uses this information to determine a configuration that maximizes an end user’s utility. This approach allows Conflux to work with many links and allows for more sophisticated configurations that include partial replication of data across multiple links.
Some preliminary results show that Conflux is able to readily adapt to changing network conditions.