Modeling window based congestion control protocols with many flows

The Markovian analysis of telecommunication networks with several interacting connections is very difficult due to the explosion of the number of states in the model. Simulation analysis under the same conditions also fails, because of prohibitive simulation times due to the enormous amount of events that must be collected to reach statistically meaningful results. This paper presents a modeling technique based on closed queuing networks, that is suitable for the evaluation of adaptive window congestion control protocols in heavy load conditions, overcoming the problem of state dimension explosion. The technique is presented applying it to a class of protocols where the window is linearly increased when the transmission is successful, and reset to 1 when packets are lost; however, the technique is general and can be applied to almost any increase and decrease policy. Results are presented for the modeled protocol class, showing the behavior as the number of competing connections increases. Then, a modification on the protocol is presented, in order to show possible ways to ameliorate the performance and, most of all, how the modeling technique can help in evaluating protocol modifications before their deployment.