Goodput Control for Heterogeneous Data Streams

Congestion control tries to answer the question: "At what rate should a sender transmit data under current network conditions?" While answering this question is sufficient to maximize the goodput of traditional bulk data streams, emerging multimedia applications generate heterogeneous data streams where different frames have different quality of service requirements in terms of priority, deadlines, and inter-frame dependence. Consequently, the goodput of a multimedia stream depends not only on answering the above question, but also the question "Which packets should the sender transmit given its current transmission rate?" In this paper, we make the case that the "goodput control" mechanisms that answer this question should be considered as a critical component of future multimedia transport protocols. We present an objective definition of goodput at the transport layer, and show that optimizing this goodput function has exponential complexity in the general case. We then present a set of simple online packet dropping algorithms that can be used at the sender side in order to approximate the optimum within a bounded ratio, and show that our goodput control mechanisms also help to improve application level reception quality for the flow.