Performance Evaluation 55 (2004) 93–111 Performance evaluation of Westwood+ TCP congestion control  S. Mascolo ∗ , L.A. Grieco, R. Ferorelli, P. Camarda, G. Piscitelli Dipartimento di Elettrotecnica ed Elettronica, Politecnico di Bari, Via Orabona 4, 70125 Bari, Italy Abstract Westwood+ TCP is a sender-side only modification of the classic Tahoe/Reno TCP that has been recently proposed to improve fairness and efficiency of TCP. The key idea of Westwood+ TCP is to perform an end-to-end estimate of the bandwidth available for a TCP connection by properly counting and filtering the stream of ACK packets. This estimate is used to adaptively decrease the congestion window and slow-start threshold after a congestion episode. In this way, Westwood+ TCP substitutes the classic multiplicative decrease paradigm with the adaptive decrease paradigm. In this paper we report experimental results that have been obtained running Linux 2.2.20 implementations of Westwood+, Westwood and Reno TCP to ftp data over an emulated WAN and over Internet connections spanning continental and intercontinental distances. In particular, collected measurements show that the bandwidth estimation algorithm employed by Westwood+ nicely tracks the available bandwidth, whereas the TCP Westwood bandwidth estimation algorithm greatly overestimates the available bandwidth because of ACK compression. Live Internet measurements also show that Westwood+ TCP improves the goodput w.r.t. TCP Reno. Finally, computer simulations using ns-2 have been developed to test Westwood, Westwood+ and Reno in controlled scenarios. These simulations show that Westwood+ improves fairness and goodput w.r.t. Reno. Keywords: TCP congestion control; Network emulation; Network measurement; Network simulation 1. Introduction Due to the fundamental end-to-end principle [1–3], classic Tahoe/Reno TCP congestion control im- plements a self-clocked additive increase multiplicative decrease (AIMD) sliding window algorithm to probe the network bandwidth capacity [4–7]. It employs two control variables: (1) the congestion window (cwnd), which limits the number of segments transmitted by the source and not yet acknowledged by the receiver (i.e. the outstanding segments), and (2) the slow-start threshold (ssthresh), which varies the way the cwnd is increased. In order to probe the network capacity, the cwnd is increased until a congestion episode happens to indicate that the network capacity has been hit. The way the cwnd is increased consists of two phases: (1) the slow-start phase, which exponentially increases the cwnd when cwnd < ssthresh,  This is an extended version of the paper “Live Internet Measurements Using Westwood+ TCP Congestion Control”, IEEE Globecom 2002 conference. ∗ Corresponding author. Tel.: +39-080-5460621; fax: +39-080-5347470. E-mail addresses: mascolo@poliba.it (S. Mascolo), a.grieco@poliba.it (L.A. Grieco), ferorelli@poliba.it (R. Ferorelli), camarda@poliba.it (P. Camarda), piscitel@poliba.it (G. Piscitelli). 0166-5316/$ – see front matter © 2003 Elsevier B.V. All rights reserved. doi:10.1016/S0166-5316(03)00098-1 Preprint Version