On the Delay Performance of Browser-based Interactive TCP Free-viewpoint Streaming

In free-viewpoint video arbitrary views of a scene or an object are rendered from a 3-dimensional scene representation that is obtained using multiple cameras or generated by computer graphics. The interactivity that is due to the viewpoint selection is particularly challenging in case of networked applications, where a server renders the scene from a viewpoint that is chosen by a remote client. Relying on widely-used standard browser-based video streaming technology, data transport is performed by the Transmission Control Protocol (TCP), implying an anticipated risk of potentially large delays. The magnitude, frequency, and origin of such delays are the focus of this work. To investigate the tail distribution of the delays, we use a controlled testbed environment and instrument the entire video streaming chain from the server-side renderer to the display at the client using various measurement points. We identify three major sources of delays: the video coders, the protocol stack, and the network. We investigate the causes of these delays and show a strong impact of network parameters, such as round-trip time and packet loss probability, on protocol stack delays. While stack delays can significantly exceed network delays, we find that stack delays can be reduced effectively by adapting the parameters of the video encoder.