IEEE 802.11b WLAN Performance with Variable Transmission Rates: In View of High Level Throughput Namgi Kim 1, Sunwoong Choi 2, and Hyunsoo Yoon 1, 1 Div. of Computer Science, Dept. of EECS, KAIST, 373-1 Kuseong, Yuseong, Daejeon, 305-701, Korea {ngkim, hyoon}@camars.kaist.ac.kr 2 School of CSE, SNU, San 56-1, Sinlim, Gwanak, Seoul, 151-742, Korea schoi@popeye.snu.ac.kr Abstract. Wireless networks have been rapidly integrated with the wired Internet and have been widely deployed. In particular, IEEE 802.11b WLAN is the most widespread wireless network today. The IEEE 802.11b WLAN supports multiple transmission rates and the rate is chosen in an adaptive manner by an auto rate control algorithm. This auto rate control algorithm deeply affects the total system performance of the IEEE 802.11b WLAN. In this paper, we examine the WLAN performance with regard to the auto rate control algorithm, especially the ARF scheme, which is the most popular auto rate control algorithm in 802.11b based WLAN products. The experimental results indicate that the ARF scheme works well in the face of signal noise due to node location. However, the ARF scheme severely degrades system performance when multiple nodes contend to obtain the wireless channel and the packet is lost due to signal collision. 1 Introduction Recently, WLAN (Wireless LAN) has achieved tremendous growth and has become the prevailing technology for wireless access for mobile devices. WLAN has been rapidly integrated with the wired Internet and has been deployed in offices, universities, and even public areas. With this trend, many WLAN technologies based on different physical frequency bands, modulations, and channel coding schemes have been proposed and implemented. In particular, the IEEE 802.11b [1] is cited as the most popular WLAN technology today. The IEEE 802.11a [2] and IEEE 802.11g [3] WLAN have better performance than 802.11b. However, IEEE 802.11b based WLAN products will not disappear for some time, given that they have already been widely deployed throughout the world. Contrary to the wired network, wireless channel conditions dynamically change over time and space. To cope with variable channel conditions, the IEEE 802.11b WLAN This work was supported by the Korea Science and Engineering Foundation (KOSEF) through the advanced Information Technology Research Center (AITrc) and University IT Research Center Project. P. Lorenz and P. Dini (Eds.): ICN 2005, LNCS 3421, pp. 1080 1087, 2005. Springer-Verlag Berlin Heidelberg 2005
IEEE 802.11b WLAN Performance with Variable Transmission Rates 1081 specification provides multiple transmission rates that can maximize the system throughput in the face of radio signal diversity. The transmission rates should be chosen adaptively depending on the conditions of the wireless channel. In general, the transmission rates are automatically changed based on the historical results of previous transmissions and this auto rate control algorithm deeply affects the WLAN performance. Therefore, in this paper, we measure the WLAN performance with variable transmission rates in IEEE 802.11b based products and show the effects of the auto rate control algorithm. 2 IEEE 802.11b WLAN The IEEE 802.11 WLAN standard [4] defines a single MAC (Medium Access Control) and a few different PHY (Physical Layer) specifications. The MAC specification deals with the framing operation, interaction with a wired network backbone, and interoperation with different physical layers. On the other hand, the PHY specifications deal with radio characteristics, modulations, error correcting codes, physical layer convergence, and other signaling related issues. The IEEE 802.11b is a kind of IEEE 802.11 WLAN standard with a particular PHY specification. In the IEEE 802.11 WLAN standard, the MAC protocol provides two different access methods for fair access of a shared wireless medium: PCF (Point Coordination Function) and DCF (Distributed Coordination Function). PCF provides contentionfree access and DCF provides contention-based access. PCF has been proposed to guarantee a time-bounded service via arbitration by a point coordinator in the access point. However, it is an optional access method and is rarely implemented in currently available 802.11 devices. In the IEEE 802.11 WLAN standard, the DCF is the mandatory and prevailing access method in the market today. Therefore, we also consider the DCF method of the IEEE 802.11 MAC protocol in this paper. In the IEEE 802.11 WLAN standards, there are various PHY specifications that provide different transmission rates by employing different frequency bands, modulations, and channel coding schemes. The 802.11b PHY [1] provides four different transmission rates ranging from 1Mbps to 11Mbps at the 2.4 GHz ISM (Industrial, Scientific and Medical) frequency band. Another PHY specification, the 802.11a PHY [2], provides high-speed transmission with eight different rates ranging from 6Mbps to 54Mbps in the 5 GHz U-NIII (Unlicensed National Information Infrastructure) frequency band. It boosts up transmission rates by adapting the OFDM (Orthogonal Frequency Division Multiplexing) technique in the physical layer. Lastly, 802.11g PHY [3] has also been developed to adapt the OFDM in the 2.4 GHz ISM band. The 802.11g PHY theoretically has the same transmission rates as the 802.11a PHY. However, it covers a larger range than 802.11a because of operation in the lower frequency band. Currently, products based on 802.11a PHY or 802.11g PHY are emerging in the consumer market. In particular, products based on 802.11g PHY have the potential to make inroads into the WLAN market because 802.11g PHY is backward compatible with 802.11b PHY. However, the most popular products in the current market are still based on 802.11b PHY. This may remain the status for some time since many 802.11b WLAN products have already been widely deployed. Therefore, we also use the 802.11b WLAN product for our experiments.
1082 N. Kim, S. Choi, and H. Yoon 2.1 Auto Rate Control Algorithm with IEEE 802.11 WLAN As noted earlier, the IEEE 802.11 WLAN standard supports multiple transmission rates. The standard, however, does not specify how to change the rates according to channel conditions. In the current WLAN fields, different auto rate control algorithms have been proposed. However, the ARF (Auto Rate Fallback) scheme [5] is the most popular auto rate control algorithm in IEEE 802.11b based WLAN products today. In the ARF scheme, the transmission rate is downgraded to the next lower rate when the transmission continually fails and as a result the ACK (acknowledgement) from the receiver is consecutively missed. The transmission rate is upgraded back to the next higher rate if the next consecutive transmissions are successful or some amount of time has passed. The ARF scheme is simple and easy to implement. However, it degrades WLAN throughput when many nodes attempt to transmit data in contention. In the next section, we show the results of our experiments and analyze the effects of the ARF scheme in IEEE 802.11b WLAN environments. 3 WLAN Performance with Variable Transmission Rates To measure WLAN performance, we conducted experiments in real IEEE 802.11b based WLAN environments. For the experiments, we used Enterasys's RoamAbout 802.11b WLAN products [6], which adopt the Agere WLAN chipset [7]. These products also adopt the ARF scheme. In these products, the ARF scheme downshifts the transmission rate after missing two consecutive ACKs and upshifts after receiving five successive ACKs. Like other venders' products, Enterasys's WLAN cards support only a contention-based DCF MAC access method. We did not use the CTS/RTS mechanism in our experiments, as is the case in many products' default setting. As mobile nodes, we used four laptops and fifteen ipaq PDAs. RedHat Linux (kernel version 2.4.7-10) [8] is installed in laptops and LinuxGPE (kernel version 0.7.2) [9] is installed in PDAs. Experiments were conducted in a building. The building has many offices and a long corridor. For each experiment, the mobile nodes send data to the wired nodes via the WLAN AP (Access Point) or vice versa. The size of a packet is 1KByte and each experiment lasted more than 300 seconds. 3.1 Throughput with Node Locations First, we measured WLAN performance in different locations with one mobile node. Fig. 1 illustrates the locations of the node and the AP. The mobile node was located sequentially from PT1 to PT5. PT1 and PT2 are located in the same room with the AP and the other three locations are out of the room in the corridor. In this experiment, a mobile node sends CBR data by UDP packets to the wired node via the AP. The offered traffic is 6.5Mb/s for each experiment. Since only one mobile node attempts to transmit data to the fixed AP in the air, the signal noise depending on the location of the mobile node becomes the most important factor in terms of signal quality. Fig. 2 shows the total throughput results corresponding to node locations. The bar indicates the throughput depending on the transmission rates at the each location from PT1 to PT5. The error bar denotes the standard deviation of observed throughput. The results indicate that the high transmission rate shows good throughput and small
IEEE 802.11b WLAN Performance with Variable Transmission Rates 1083 Fig. 1. Locations of mobile nodes and AP Fig. 2. Throughputs with node locations variation over the near locations. However, as the distance between the location and the AP increases, the high transmission rate suffers from low throughput and large variation because of radio signal noise. The low transmission rate, on the other hand, is relatively robust against signal noise and shows stable performance at all locations. The low transmission rate, however, cannot maximize throughput when the channel condition is good at a near location. As you can see in the results, the 11M auto rate shows relatively good performance in comparison to the fixed transmission rates. The 11M auto rate displays high throughput and low throughput variation over all locations. This is because the ARF scheme works well with packet loss due to signal noise. When only one node attempts to send data, there is no contention and the signal noise depending on location becomes the most important factor in terms of signal quality. Consequently, the 11M auto rate, which adapts the ARF scheme, works well over all locations. We also conducted the experiments with a downlink traffic pattern. Thus, the wired node sends data to the mobile node. The results of the experiments with downlink traffic are very similar to those obtained from the previous uplink traffic experiments and are thus not presented. 3.2 Throughput with Node Contentions In this experiment, we measured the effects of node contentions in WLAN environments. Multiple mobile nodes send CBR data through the AP to the wired nodes in a position. The nodes are located in the same place and contend to obtain the wireless channel. The total offered traffic is 6.5Mb/s and the number of nodes is changed for each experiment. Fig. 3 shows total throughputs corresponding to different numbers of nodes. In the results, we find that the fixed transmission rates are rarely affected by the number of mobile nodes and they show good performance regardless of the number of contending nodes. However, the throughput of the 11M auto rate is severely degraded when the number of nodes is increased. When the number of the mobile nodes is eighteen, the throughput of the 11M auto rate is almost the same as that of the 1M fixed transmission rate.
1084 N. Kim, S. Choi, and H. Yoon Fig. 3. Throughputs with node contentions Fig. 4. Channel occupancy at 18 nodes To analyze this phenomenon, we captured the WLAN frames using the Airopeek tool [10]. Through this trace, we examined the channel occupancy according to transmission rates. Fig. 4 shows the ratio of the channel occupation time when the 11M auto transmission rate is employed to eighteen mobile nodes. The results reveal that the 1M rate, the lowest transmission rate, occupies the channel for the greatest period of time and the 5.5M rate, the next lower rate, is second. Consequently, we determined that the throughput of the 11M auto rate was drastically degraded, because the low transmission rates occupied the channel for a lengthy period of time. According to the results, the ARF scheme does not work correctly in the face of node contentions. When the number of nodes is increased and the nodes contend to obtain a wireless channel in the 802.11 DCF MAC protocol, the packet is lost due to collisions in the air. Packet loss due to collision causes the transmission rate to downshift unnecessarily even though the signal quality of the wireless channel is good. Moreover, this loss obstructs quick upshifting back to higher transmission rates. Therefore, the throughput with the auto rate control algorithm based on the ARF scheme is severely degraded when the node contention is heavy, and consequently packet loss due to signal collision occurs frequently. 3.3 Throughput with TCP In this experiment, we measured the WLAN performance with TCP in node contentions. To do this, we send TCP data through the AP to the eighteen mobile nodes. TCP is a bidirectional protocol. Thus, the mobile nodes contend to obtain the wireless channel to send back TCP ACK packets. The other experimental setups are the same as those of the previous node contention experiments with UDP packets for CBR traffic. Fig. 5 shows the TCP throughput corresponding to different transmission rates with eighteen mobile nodes. The throughputs of the fixed transmission rates are similar to the results of the previous CBR experiments. However, the throughput of the 11M auto rate is quite different. The throughput of the 11M auto rate is not severely degraded. It is almost the same as that of the 11M fixed transmission rate, even though the eighteen mobile nodes contend with each other to obtain the wireless channel. To analyze this effect, we captured the WLAN frames again. Using the raw WLAN trace, we calculated the active nodes attempting to obtain the channel at the same time. In these TCP experiments, the active node is the actual node that is involved in
IEEE 802.11b WLAN Performance with Variable Transmission Rates 1085 Fig. 5. TCP throughput with 18 nodes Fig. 6. Active nodes in contentions contention. We define a node as active if the AP has sent a TCP DATA packet to the mobile node, but the mobile node has yet to send back the TCP ACK packet to the AP. This definition is not strictly correct because the TCP adopts a commutative ACK mechanism. However, we believe this definition is adequate in terms of explaining the TCP performance with the ARF scheme. Fig. 6 shows the number of active nodes per second when the eighteen mobile nodes employ the 11M auto rate. The average number of active nodes is 3.64, which is quite smaller than the number of actual mobile nodes. Accordingly, we infer that the throughput of the 11M auto rate is not degraded because the contention is relieved when using TCP. TCP reduces the degree of node contentions because it has already adopted rate control mechanism. A TCP ACK packet is generated by successfully transmitted a TCP DATA packet, and a TCP DATA packet is also generated by successfully transmitted a TCP ACK packet. Thus, the mobile node cannot be involved in contentions before receiving a TCP DATA packet from the AP. Also, the AP cannot send the next packet before the mobile node sends back a TCP ACK packet to the AP. In the IEEE 802.11 WLAN, the contention-based DCF access method fairly distributes the wireless channel to all stations. The DCF method does not give greater priority to the AP, which causes to increase the active nodes by consecutive frame transmissions from the AP to every different mobile node. Therefore, when traffic is regulated by TCP, the number of contending nodes is maintained smaller than the actual nodes. The upstream TCP experiments also showed similar results. This is because TCP regulates contention of nodes attempting to send TCP DATA packets to the wired nodes through the AP reversely. 4 Conclusions and Discussion The IEEE 802.11b WLAN standard supports multiple transmission rates and the auto rate control algorithm attempts to select the best transmission rate based on the current wireless channel condition. The most popular auto rate control algorithm in the current IEEE 802.11b WLAN products is the ARF scheme. The ARF scheme changes the transmission rates based on the success or failure of previous frame transmissions. This mechanism shows good performance when few mobile nodes exist and channel quality is dominated by signal noise due to location of the node. However, if many
1086 N. Kim, S. Choi, and H. Yoon nodes attempt to send data in contention and some amount of packets is lost due to signal collision, the ARF scheme operates improperly. Packet loss due to collision causes the ARF scheme to downgrade the transmission rate inaccurately and delays upgrade back to the high rate. This is because the ARF scheme does not distinguish between signal noise and collisions. Control by protocol, such as TCP, can diminish the inappropriate effects of the ARF scheme in node contention. Since the TCP has a rate control mechanism based on TCP ACK, it decreases the number of actively contending nodes in probabilistically fair DCF access method. However, the TCP is not employed to all WLAN applications. Some applications, such as VoIP and multimedia data, adopt the UDP instead of the TCP. Therefore, the TCP cannot be an optimal solution for all WLAN applications. Performance abnormality with variable transmission rates in IEEE 802.11b WLAN has been recently noted in some papers [11, 12]. M. Heusse et al. [11] observed performance anomalies with multiple transmission rates. They pointed out that when some mobile nodes use a lower transmission rate than others, the overall WLAN system performance is considerably degraded. However, they only addressed the relationship between the system performance and the mixed transmission rates. They did not determine the effect of an auto rate control algorithm and TCP protocol. In [12], P. Berthou et al. noted that the auto rate control algorithm could degrade the system performance and it should be disabled for multimedia applications. However, they also did not determine why an auto rate control algorithm such as the ARF scheme degrades WLAN system performance. Moreover, they did not perform indepth study on this phenomenon. The newly proposed SNR-based algorithm [13] may solve this performance abnormality with the ARF scheme. The SNR-based auto rate control algorithm adaptively changes the transmission rates based on the SNR value of the received packets. It can prevent degradation of the WLAN system performance in node contentions by distinguishing between signal noise and collision. However, the SNR-based algorithm has not been applied in practice. Moreover, unresolved problems such as a means to obtain exact SNR, a practical mapping function between the rate and SNR value, and asymmetry of up and down wireless channels must still be addressed. We have evaluated the performance of IEEE 802.11b WLAN with variable transmission rates. In particular, we examined the effect of the auto rate control algorithm in real WLAN environments and revealed why the ARF scheme degrades WLAN performance in node contentions. Lastly, we determined the relationship between TCP and the auto rate control algorithm. For future work, we plan to investigate other different auto rate control algorithms with different WLAN standards such as IEEE 802.11a and 802.11g. References 1. IEEE Std 802.11b-1999, "Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications: High-speed Physical Layer Extension in the 2.4GHz Band," Supplement to ANSI/IEEE Std 802.11, Sep. 1999. 2. IEEE Std 802.11a-1999, "Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications: High-speed Physical Layer Extension in the 5GHz Band," Supplement to ANSI/IEEE Std 802.11, Sep. 1999.
IEEE 802.11b WLAN Performance with Variable Transmission Rates 1087 3. IEEE Std 802.11g-2003, "Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications: Further Higher Data Rate Extension in the 2.4GHz Band," Amendment to IEEE 802.11 Std, Jun. 2003. 4. IEEE Std 802.11, "Wireless LAN MEdium Access Control (MAC) and Physical Layer (PHY) specifications," ANSI/IEEE 802.11 Std, Aug. 1999. 5. A. Kamerman and L. Monteban, "WaveLAN-II: A High-Performance Wireless LAN for the Unlicensed Band," Bell Labs Technical Journal, pp. 118-133, Summer 1997. 6. http://www.enterasys.com/ 7. http://www.agere.com/ 8. http://www.redhat.com/ 9. http://www.handhelds.org/ 10. http://www.wildpackets.com/ 11. M. Heusse, F. Rousseau, G. Berger-Sabbatel, A. Duda, "Performance Anomaly of 802.11b," In proceedings of IEEE INFOCOM 2003, Mar. 2003. 12. P. Berthou, T. Gayraud, O. Alphand, C. Prudhommeaux, M. Diaz, "A Multimedia Architecture for 802.11b Networks," In proceedings of IEEE WCNC 2003, Mar. 2003. 13. J. P. Pavon and S. Choi, "Link Adaptation Strategy for IEEE 802.11 WLAN via Received Signal Strength Measurement," In proceedings of IEEE ICC 2003, May 2003.