Media and protocols for multimedia communications over disparate networks

It is widely recognized that multimedia applications will be a "killer-app." These applications depend so heavily on the type of network used that no real time application can claim to be truly network-independent. We present two disparate networks and address the challenges in building real-time applications for each. First we examine a network which can provide guarantees on bandwidth and delay. Despite these network guarantees, the interaction between the server and the client can severely affect the performance of the system, including scalability. We present a model for providing an end-to-end guarantee on Quality of Service (QoS) by mapping QoS parameters at the video server to subjective parameters at the video client. We then present a network which cannot guarantee QoS. This network requires special protocols and algorithms for maintaining quality for the end user. Applications operating in this environment must continually adapt to the state of the network and must do so in a way that does not damage the performance of other traffic on the network. This involves extracting implicit or explicit information about the state of the network. It also requires adapting the data (audio, video or images) so that they do not demand more resources than the network can provide. We present a variety of codecs for different media and show ways in which they can be made adaptable. We then present a new semi-reliable transport protocol for streaming real time media on today's Internet. The protocol uses the TCP congestion window to pace the delivery of data into the network, without using other TCP algorithms that are poorly suited to real time protocols. For this reason, our protocol competes fairly with all other Internet data by optimally sharing the available bandwidth. We then examine the performance of TCP, and hence the performance of our protocol, over wireless networks. A model for TCP utilization in the presence of random wireless errors is presented which is much more tractable than previous work. This model is used to draw conclusions about the utilization of TCP over wireless Local Area Networks.