Utility maximization for asynchronous streaming of bufferable information flows

We consider optimizing the streaming of bufferable information flows to multiple clients sharing a network. The information flow to each client can be broken down into time segments where each segment is associated with a possibly varying quality/distortion-rate trade-off which can be adapted to network resources available and allocated to the client. The segments are downloaded, buffered and consumed sequentially by each client, and this proceeds in an asynchronous manner across the clients. Such settings are relevant to streaming of video and audio, and potentially to streaming of augmented reality and virtual reality content. We focus on jointly optimizing the network’s resource allocation and clients’ quality adaptation across segments so as to fairly optimize clients’ Quality of Experience (QoE), while incorporating clients’ sensitivity to rebuffering events caused when a client’s buffer empties. We consider QoE models capturing trade-offs between clients’ mean quality and temporal variability in quality. We present a simple asymptotically optimal online algorithm to solve the problem. It distributes the tasks of resource allocation to the network and quality adaptation to the respective clients. Further, it is asynchronous and is lightweight in terms of implementation overheads.