An Experimental Study of Throughput for UDP and VoIP Traffic in IEEE 802.11b Networks Sachin Garg sgarg@avaya.com Avaya Labs Research Basking Ridge, NJ USA Martin Kappes mkappes@avaya.com Avaya Labs Research Basking Ridge, NJ USA Abstract—In this paper, we present experimental studies on the throughput of IEEE 802.11b wireless networks for UDP and VoIP traffic. Our experiments show that the maximum data throughput of a single station sending out UDP traffic is 6.1 Mbps. The maxi- mum number of VoIP calls in a single cell of an IEEE 802.11b net- work is six if the ITU G711a-Law codec is used with 10 millisec- onds of audio data per RTP packet. The experiments also show that the effective available bandwidth in the wireless network is reduced by ongoing VoIP connections. Specifically, for the above codec settings, each VoIP connection reduces the bandwidth avail- able for data traffic by 900 Kbps. I. I NTRODUCTION In the last few years, wireless networks based on the IEEE 802.11b standard have gained popularity and have been widely deployed in enterprises mostly to provide wireless data access from Laptops, PDAs, etc. to the wired infrastructure of the en- terprise. They have also been deployed in public hot-spots such as airports, hotels, conference facilities etc., mainly for inter- network connectivity. The maximal data rate 802.11b currently supports is 11 Mbps. Although task groups, in particular 802.11a and 802.11g [10], are working on allowing higher maximum data dates, it is very likely that 802.11b deployments will continue to op- erate in both enterprises as well as the residential market for the next few years. As opposed to IEEE 802.3 (also known as Ethernet), where the maximal data rate of the network is in- deed close to the throughput observed, the maximal achievable throughput for 802.11b networks is far lower than the data rate due to the nature of 802.11’s CSMA/CA medium access proto- col. Moreover, the bandwidth is shared among all participants in the network whereas most of today’s Ethernet deployments are “switched”. As converged networking in the wired world is becoming more and more popular, it is very likely that in the near fu- ture wireless networks will also be increasingly used for voice traffic. In this paper, we present our experimental studies on the throughput for VoIP and UDP traffic in a single cell of an 802.11b network. We chose UDP as transport layer protocol due to its connection-less nature. A UDP sender “flooding” the network gives an accurate estimate of the actual bandwidth that is avail- able in the network. Results obtained with UDP constitute an upper bound for the throughput possible with TCP as transport layer protocol in any of the scenarios we studied. Another rea- son why we chose UDP is the 802.11 MAC. It was constructed in such a way that higher frame loss and collision rates in the wireless network as compared to a wired network are mitigated by the MAC-layer itself. In particular, the MAC protocol man- dates acknowledgments of received frames by the receiver and retransmissions of frames by the sender that are not acknowl- edged within a time interval. The timer values are such that the higher layer transport protocols, in particular TCP, do not get affected by the loss of a frame once in a while. Hence, studying phenomena related to the connection-oriented nature of TCP does not lead to new insights into network problems that are related to the specific use of an 802.11 network. This paper is organized as follows. First, in Section II, we outline the laboratory setup for all conducted experiments. Each section from III through VI consists of the description of a single experiment, the results obtained from the experiment and a detailed explanation. The first experiment, in Section III, studies the payload throughput of an 802.11b network as a function of the frame payload size. The second experiment, in Section IV, studies the throughput with multiple senders. The third experiment, in Section V, is aimed to find the maximum number of simultaneous VoIP connections in a single cell. In Section VI, we study the throughput of such a network in the case of converged networking, i.e., simultaneous voice and data traffic. Finally, Section VII discusses the ramifications of the experiments and Section VIII concludes the paper. II. EXPERIMENTAL SETUP Eight clients, all PCs (some Laptops, some Desktops) run- ning Windows 2000, are associated with a single access point. The access point is connected to an IEEE 802.3 (Ethernet) LAN. On the wired side of the network, we are also using PCs running Windows 2000 which serve as endpoints for VoIP or UDP data connections. All endpoints in the wireless network are in one subnet whereas the nodes in the wired network are on another. Both subnets are connected through a single router and are thus one hop apart from each other. The IEEE 802.11b access point and all clients are situated in the same room with no physical obstacles between them. Hence, the probability of frame loss due to weak signal strength and/or presence of hidden stations is negligible. The experi- ments were conducted with access points from different ven- dors, with no significant differences in the results. The radio cards used in all clients were from the same vendor. 1748