Experience with control mechanisms for packet video in the internet

Abstract

The single class best effort service available in the current Internet does not provide the guarantees, typically expressed in terms of minimum bandwidth and/or maximum delay or loss, associated with real-time applications such as live video. One way to support such applications in best effort networks is to use control mechanisms that adapt the coding, transmission, reception, and decoding processes at the source and at the destination(s) depending on the state of the network. In this paper, we examine and report on our experience over the past several years with such mechanisms for videoconferencing software. We illustrate our points with results obtained with the IVS software developed at INRIA.We consider in particular rate and error control mechanisms. These mechanisms adapt the bandwidth requirements and the resilience to packet loss of the video stream sent by a source coder. We have found that they do prevent video sources from swamping the resources of the Internet, and that they provide reasonably graceful degradation of image quality during periods of network congestion. However, they raise thorny issues of fairness with non-real time traffic sources (such as TCP sources) and of scalability for the multicast delivery of video in heterogeneous networks. We examine different approaches to tackling these problems.