Stochastic Petri Nets models for the performance analysis of TCP connections supporting finite data transfer

This paper proposes the Stochastic Petri Nets (SPN) formalism to model IP networks loaded with traffic resulting from a set of ON-OFF finite TCP-Reno connections. The approach is based on separate model descriptions of the TCP connection latency to transfer a finite number of packets and the IP network links; the two models are iteratively solved using a fixed point method. The overall model parameters are the primitive network characteristics; the model solution yields an estimation of the packet loss probability, the average completion time to transfer a finite number of packets over a TCP connection as well as the distribution of the completion time. The validation of the proposed approach and the future extensions to this work will also be discussed.