An Analytical Model for Streaming over TCP

Streaming over TCP has become popular as demonstrated by the example of YouTube. To cope with variability in data throughput, streaming applications typically implement buffers. Yet, for improving the quality of user experience, it is critical to dimension buffers and initial buffering delays appropriately. In this paper, we develop an analytical framework that describes the dimensioning of appropriate buffers. To this end, we propose to rely on modeling congestion window sizes immediately before a triple duplicate or timeout event. We observe that such “bounds” on TCP window sizes follow a Gamma distribution. Although being of general use due to its simplicity and accuracy, our proposed TCP model is particularly useful for TCP streaming. As confirmed by experiments, it allows to estimate the frequency of buffer overflow or underflow events if buffer sizes and initial buffering delays are known parameters in the proposed TCP streaming model, or conversely, to dimension the buffer appropriately.