Disambiguating network effects from edge effects in TCP connections

While the applications using the Internet have changed over time, TCP is still the transport protocol that is used by almost all the applications and carries over 90% of the total traffic. Throughput is a key performance metric for long TCP connections. The achieved throughput results from the aggregate effects of the network path, the TCP end points’ parameters, and the application on top of TCP. Finding out which of these is limiting the throughput of a TCP connection is a critical problem for end users that want to understand the origins of their problems, ISPs that need to troubleshoot their network, and application designers that need to know where the bottleneck is. In this paper, we revisit this issue by first demonstrating the weaknesses of the previously proposed flight-based approach. We next discuss in detail the different possible limitations and highlight the need to account for the application behavior during the analysis process. This paper’s main contribution is a new approach based on the analysis of time series extracted from packet traces that allow for quantitative assessment of the different causes to the resulting throughput. We exemplify the interest of this approach on a large BitTorrent dataset.