Maximizing video quality for several unicast streams in a multipath overlay network

A streaming system that uses an overlay network for multipath streaming needs to make decisions concerning the distribution of the available bandwidth among all of its clients. This decision making should aim at delivering the best possible quality to all clients while providing an optimal utilization of the network resources. We consider a scenario where videos are hierarchically layered-encoded and most requests are negligibly overlapped in time. It implies that using multicast is not efficient, and instead, the streams are striped and allocated to multiple paths from the server to the client. To evaluate how well the rate-allocation algorithms approach optimality, we have earlier built a benchmarking system that provides the optimal solution for assigning available bandwidth to delivery paths. However, as video quality is not linearly related to bitrate, the trivial maximization of the total consumed bandwidth does not necessarily maximize the video quality. To address this problem, we define a metric that assesses video quality for a group of clients that we use as a utility function in the revised benchmarking system. Due to its concavity, this utility function distributes the bandwidth resources proportionally fair between the clients of the system.