TCP-SACK Analysis and Improvement through OMQN Models.

Protocol design and modification was traditionally based on heuristic processes and decisions. In particular several new versions of TCP have been proposed which not always stand up to their expectations when they are deployed or simply simulated in a real scenario. The closed-loop, nonlinear characteristics of TCP makes it very difficult to predict its performance in even simple scenarios. This papers analyses the process of modeling TCP-SACK with an Open Multiclass Queueing Network (OMQN) model, showing that sometimes simple implementation choices can have a non-marginal impact on the performance and discussing how an analytical model can be used to design and study TCP (or other protocols) modifications and improvements. As an example we present a modified, more robust implementation of TCP-SACK, named R-SACK, that significantly reduces the protocol timeout probability in presence of bursty losses. [ABSTRACT FROM AUTHOR]