An Algorithm for the Evaluation of the Throughput of a TCP NewReno Bulk Data Flow

Abstract

This letter presents a simple and accurate algorithm to evaluate the steady-state throughput of a TCP NewReno bulk data flow as a function of the round-trip time, the retransmission timer value, and the loss behavior of the network. This work proposes a new method for evaluating the throughput of a TCP NewReno connection over any network whose end-to-end packetloss process can be described by an analytical model or by any other method that results in the sequence of lost packets along the transmission. This proposal achieves more accurate results than those corresponding to the existing analytical models, with greater simplicity and less resource consumption than those necessary in a simulation. Such an algorithm is validated by its comparison with a set of ns-2 simulation results, obtaining an average error of 1.4% when the loss behavior of the network is modeled by a Bernoulli process (independent network losses) and of 9% when the Gilbert loss model (correlated network losses) is used.