Networking 2006: Networking Technologies, Services, and Protocols; Performance of Computer and Communication Networks; Mobile and Wireless Communications Systems
IFIP Networking 2006
15-19 May 2006
[en] TCP ; Machine Learning ; Throughput model
[en] Based on a large set of TCP sessions we first study the accuracy of two well-known analytical models (SQRT and PFTK) of the TCP average rate. This study shows that these models are far from being accurate on average. Actually, our simulations show that 70% of their predictions exceed the boundaries of TCP-Friendliness, thus questioning their use in the design of new TCP-Friendly transport protocols. Our study also shows that the inaccuracy of the PFTK model is largely due to its inability to make the distinction between the two packet loss detection methods used by TCP: triple duplicate acknowledgments or timeout expirations. We then use supervised learning techniques to infer models of the TCP rate. These models show important accuracy improvements when they take into account the two types of losses. This suggests that analytical model of TCP throughput should certainly benefit from the incorporation of the timeout loss rate.
Fonds de la Recherche Scientifique (Communauté française de Belgique) - F.R.S.-FNRS