Additive Increase Adaptive Decrease Congestion Control a Mathematical Model and Its Experim.pdf

IEEE ISCC July 1-4 2002, Giardini Naxos, Italy Additive Increase Adaptive Decrease Congestion Control: a Mathematical Model and Its Experimental Validation1 1 This work was partially supported by the MIUR Research Project 488/92 C22 entitled e-service to the Citizens L.A. Grieco Dipartimento d’Ingegneria dell’Innovazione Università di Lecce, Italy alfredo.grieco@unile.it S. Mascolo Dipartimento di Elettrotecnica ed Elettronica Politecnico di Bari, Italy mascolo@poliba.it R. Ferorelli Dipartimento di Elettrotecnica ed Elettronica Politecnico di Bari, Italy Abstract Due to the fundamental end-to-end design principle of the TCP/IP for which the network cannot supply any explicit feedback, today TCP congestion control algorithm implements an additive increase multiplicative decrease (AIMD) algorithm. It is widely recognized that the AIMD mechanism is at the core of the stability of end-to-end congestion control. In this paper we describe a new mechanism we call Additive Increase Adaptive Decrease (AIAD). The key concept of the adaptive decrease mechanism is to adapt congestion window reductions to the bandwidth available at the time the congestion is experienced. We propose Westwood++ TCP as an implementation of the AIAD paradigm, and we consider Reno TCP as an example of the AIMD mechanism for comparison. We derive a mathematical model of the throughput of the AIAD mechanism that shows that Westwood++ is stable, is friendly to Reno and increases the fairness in bandwidth utilization. To confirm the validity of the theoretical model Internet measurements are reported. 1. Introduction The stability of the Internet requires that flows use some form of end-to-end congestion control to adapt the input rate to the available bandwidth. TCP Congestion control was introduced into the Internet in the late 1980s by Van Jacobson after that the Internet had experienced conge