A Queue Management Mechanism for Improving TCP fairness in Wireless Access Networks Pedro Henrique Gomes 1 , Dzmitry Kliazovich 2 , Fabrizio Granelli 2 , and Nelson L. S. da Fonseca 1 1 Institute of Computing – State University of Campinas Av. Albert Einstein, 1251 – Campinas - Brazil psilva@lrc.ic.unicamp.br, nfonseca@ic.unicamp.br 2 DISI – University of Trento Via Sommarive 12, I-38050 – Trento - Italy {kliazovich,granelli}@disi.unitn.it Abstract. Due to the contention nature of Wi-Fi medium access control, several fairness issues exist. Moreover, the Access Point (AP) buffer space plays an important role in the provisioning of fairness to the flows. This paper presents a novel queue management technique designed to ameliorate fairness problems in Wi-Fi last mile networks. The proposed mechanism allocates the AP buffer space based on information available at the mobile nodes and sent to AP via link layer acknowledgement frames. The proposed mechanism is transparent to higher layers as well to the wired part of the network. Performance results show that the novel mechanism leads to considerable improvement when compared to traditional drop-tail queue management. 1. Introduction The IEEE 802.11 (Wi-Fi) infrastructure mode consists of two types of nodes: access point (AP) and mobile station (MS). The AP connects the wired and the wireless parts of the network, allowing the MSs to communicate with fixed stations and servers on the Internet. The IEEE 802.11 standard does not differentiate the operation of AP and MSs for the access to the medium and for packets buffering. As a result, the APs become the bottlenecks for end-to-end connections, and their buffer key points for enhancing the network performance. Despite the emergence of real-time and peer-to-peer application, mostly based on User Datagram Protocol (UDP), the Transmission Control Protocol (TCP) is still responsible for the majority of the bytes transferred on the Internet [Williamson 2001]. The throughput of TCP connections is controlled by the congestion control mechanism, which evolution is driven by the pattern of acknowledgements generated by the TCP receiver. Flow and congestion control mechanisms control the amount of data with pending acknowledgement the TCP sender is allowed to send into the network. In steady state, TCP increases its transmission rate until a segment is lost. If the loss is due to the reception of three duplicate acknowledgements, the TCP sender reduces its sending window by half. This mechanism is intended to promote equal share of the available bandwidth among TCP connections using the same link. However, such fairness cannot be guaranteed when a part of the flows receive acknowledgements from XXVIII Simpósio Brasileiro de Redes de Computadores e Sistemas Distribuídos 871